h01 log

2026-04-27 (wetlab handoff packet v0.1 delivered)

H01 WETLAB_HANDOFF_PACKET_READY_FOR_RFQ. Filed a private assay-scoping packet that reframes H01 as target-engagement feasibility, not lead-candidate nomination. Package includes a compound sheet (1-indanyl_acylsulfonamide_SO2Me_-Cl affinity track, adamantyl_acylsulfonamide_SO2Me_-Cl practical/ADMET backup, optional adamantyl_-CF3 residence comparator), a gap matrix, and an RFQ email template. P0 empirical gaps are physical compounds, recombinant STRC K1141-fragment reagent, and direct binding/stabilization data. Ranking delta: A held; mech 4 / deliv 3 / misha_fit 4 unchanged. Next step changes to send separate RFQs for custom synthesis, STRC 1066-1216 WT/E1659A fragment expression, and DSF/nanoDSF + SPR/BLI/MST assay-scoping; production holo MD/MM-PBSA is paper-grade support, not a wetlab-contact blocker. → STRC h01 Wetlab Handoff Packet 2026-04-27

2026-04-27 (14:38 — Phase 9b min-relax multi-frame MM-GBSA smoke delivered)

Phase 9b MINRELAX_MULTIFRAME_SMOKE_SUPPORTIVE. Reused the repaired matched WT/E1659A Amber dry complexes for v5.3__aq3__adamantyl__acylsulfonamide_SO2Me__-Cl, ran 60 extra implicit-GB minimization steps with protein :1-1775 restrained at 10 kcal/mol/A^2 and ligand free, wrote 6 frames per condition, and rescored with MMPBSA.py GB (igb=5, saltcon=0.150). WT DELTA TOTAL -15.84 +/- 0.46 SEM kcal/mol; E1659A -18.18 +/- 0.09; E1659A-minus-WT -2.35 kcal/mol. Ranking delta: A held; mech 4 / deliv 3 / misha_fit 4 unchanged. P0 next: production explicit-solvent holo MD / MM-PBSA if Phase 9b is to become paper-grade, otherwise keep this as a caveated robustness smoke. → STRC h01 Phase 9b Min-Relax Multi-Frame MM-GBSA Smoke 2026-04-27

2026-04-27 (14:13 — Phase 9b repaired matched Amber inputs + minimized endpoint smoke delivered)

Phase 9b GEOMETRY_REPAIRED_MINIMIZED_ENDPOINT_SMOKE_PASS_WITH_CAVEATS. The old blocker was traced to the Amber input path, not to unfinished AlphaFold output: Phase 5q τRAMD ligand construction preserved the Vina pose centroid but not the full docked orientation, and the first Phase 9b builder also left LIG.mol2 effectively neutral. Rebuilt the matched WT/E1659A snap_010 inputs from the full-length Phase 5e-v2 Vina PDBQT pose for v5.3__aq3__adamantyl__acylsulfonamide_SO2Me__-Cl, removed Meeko dummy atoms for OpenBabel, converted to SDF, then patched formal-charge Meeko/Vina PDBQT charges into GAFF2 mol2; WT and E1659A ligand charge sums are both -1.000000. tleap now PASSes for both conditions. Raw docked-pose geometry is much improved (WT 0 all contacts <1.0 A, one heavy-heavy contact <1.2 A at LIG O1 — GLN1144 OE1 1.167 A; E1659A 0 heavy-heavy contacts <2.5 A), but raw WT MM-GBSA remains invalid because VDWAALS is 76929.6 kcal/mol. After 80-step restrained minimization, both WT and E1659A have 0 heavy-heavy contacts <2.5 A and finite one-frame MM-GBSA smoke values: WT DELTA TOTAL -13.03 kcal/mol, E1659A -17.79 kcal/mol, E1659A-minus-WT -4.77 kcal/mol. Ranking delta: A held; mech 4 / deliv 3 / misha_fit 4 unchanged. P0 next: promote from compatibility smoke to paper-grade evidence by running multi-frame relaxed sampling or explicitly confine this to a smoke figure caveat. → phase9b_geometry_sanity

2026-04-27 (13:31 — Phase 9b geometry sanity delivered: matched complexes are physically clashed)

Phase 9b geometry sanity INPUT_GEOMETRY_CLASHED. Parsed WT/E1659A complex_dry.pdb files from matched Amber inputs. WT has 23 all-atom contacts <1.0 A and 56 heavy-heavy contacts <2.0 A; worst heavy contacts are LIG C16 — PRO738 O 0.607 A, LIG C11 — LEU742 CA 0.643 A, LIG N2 — GLN743 N 0.677 A. E1659A has 18 all-atom contacts <1.0 A and 29 heavy-heavy contacts <2.0 A; worst heavy contacts are LIG N2 — VAL731 CG1 0.693 A, LIG C5 — VAL731 CG1 0.755 A, LIG O3 — LEU735 CG/CD2 0.847/0.852 A. LIG.mol2 charge sums are neutral (WT -0.001000, E1659A 0.000001) while the v5.3 sulfonamide design previously implied monoanion state, so charge state also needs confirmation before relaxation. Ranking delta: A held; mech 4 / deliv 3 / misha_fit 4 unchanged. P0 next: rebuild/relax matched holo inputs with charge state corrected or explicitly justified, remove local overlaps, then repeat 1-5 frame endpoint smoke. → phase9b_geometry_sanity

2026-04-27 (13:12 — Phase 9b matched Amber inputs + one-frame MMPBSA.py read delivered)

Phase 9b input compatibility INPUTS_READY_ENDPOINT_READ_PASS_NUMBERS_INVALID. Builder produced matched WT and E1659A dry Amber complexes for v5.3__aq3__adamantyl__acylsulfonamide_SO2Me__-Cl on snap_010: both tleap runs PASS with Errors 0 / Warnings 4; residue sanity WT K1141=LYS/E1659=GLU, MUT K1141=LYS/E1659=ALA; ring RMSD WT 0.877 A, MUT 0.000 A. One-frame MMPBSA.py GB smoke reads both endpoint sets and finishes normally (WT 2.120 min, E1659A 1.863 min), so the prior matched-input blocker is closed. Do not use the one-frame DELTA TOTAL values as science: WT 1.705e10 and MUT 1.308e8 kcal/mol are nonphysical, driven by huge VDWAALS terms. Ranking delta: A held; mech 4 / deliv 3 / misha_fit 4 unchanged. P0 next: minimize/equilibrate matched holo inputs, inspect clashes and ligand charge state, then rerun endpoint smoke on relaxed frames before any WT-vs-MUT MM-PBSA sign. → STRC h01 Phase 9b Matched Amber Input Compatibility 2026-04-27

2026-04-27 (11:54 — Phase 9b MM-PBSA tooling preflight delivered: software ready, matched inputs blocked)

Phase 9b preflight TOOLING_READY_INPUTS_BLOCKED. Installed and verified local MM-PBSA path: Homebrew GROMACS 2026.1, gmx_MMPBSA 1.6.4, AmberTools 23.3 (MMPBSA.py, cpptraj, tleap, antechamber) in /opt/miniconda3/envs/strc-mmpbsa with AMBERHOME set. Input audit found no native h01 .top/.tpr/.xtc/.trr/.gro/.prmtop/.inpcrd; Phase 5q has E1659A OpenMM holo assets for the v5.3 lead but no matched WT holo counterpart. Attempted ParmEd OpenMM→Amber conversion loads 56,785 atoms / 18,337 residues but cannot write valid Amber topology because 37,405 bonds lack Amber bond types. No WT-vs-E1659A MM-PBSA sign produced. Ranking delta: A held; mech 4 / deliv 3 / misha_fit 4 unchanged. P0 next: build matched WT and E1659A holo systems through one Amber/GROMACS-compatible topology path, then rerun 9b smoke. → STRC h01 Phase 9b MM-PBSA Tooling Preflight 2026-04-27

2026-04-27 (11:29 — Phase 9b-lite residue-field decomposition smoke delivered: broad E1659A>WT electrostatic sign survives APBS-free post-processing)

Phase 9b-lite SMOKE_PASS. Existing matched n=20 WT + n=20 E1659A MD snapshots post-processed with formal side-chain q/r around the K1141-ring centroid; broad shells reproduce the E1659A>WT electrostatic sign (18 A delta +0.181 q/A, paired p=1.54e-10; 12-24 A positive) while the 9 A inner shell opposes (delta -0.0849 q/A). Supports Phase 5k APBS as a wider charged-neighbourhood effect, not a uniformly positive microenvironment. Ranking delta: A held; mech 4 / deliv 3 / misha_fit 4 unchanged; Phase 9b production MM-PBSA still pending gmx_MMPBSA or AmberTools path. → STRC h01 Phase 9b-lite Residue Field Decomposition 2026-04-27
Phase 9x SUPPORTIVE. v5.3 lead-committee alpha sensitivity delivered on existing Boltz-2 + Vina n=20 + tauRAMD n=20 + ADMET-AI data. Rank model = mean_affinity_rank + alpha*mean_kinetic_rank + mean_admet_rank; adamantyl_acylsulfonamide_SO2Me_-Cl remains rank-1 for every alpha in 0.5-2.0, ahead of CF3 and 1-indanyl. Supports practical lead robustness; no mechanism/delivery score move. Ranking delta: A held; mech 4 / deliv 3 / misha_fit 4 unchanged. → STRC h01 Phase 9x v5.3 Lead Committee Alpha Sensitivity 2026-04-27

Phase 9 plan PLAN. Five orthogonal in-silico tracks scoped as runnable skeletons before launch. Each track has: question, method, smoke test (1-day), production protocol, pass criteria, known artifacts, canonical references. No production runs yet. Pass for the phase as a whole = ≥3 of 5 tracks reproduce the differential sign. → STRC h01 Phase 9 Orthogonal Cross-Checks Plan 2026-04-27
Phase 9a SKELETON. FEP/TI alchemical ΔΔG (E→A on protein side, λ-windows). Tooling: GROMACS-FEP+pmx or AMBER+pmx. Charge-changing mutation needs Rocklin finite-size correction. → STRC h01 Phase 9a FEP-TI E1659A Alchemical 2026-04-27 SKELETON
Phase 9b SKELETON. Holo MD 200ns × 3 replicas + MM-PBSA rescore + per-residue decomposition. Tooling: GROMACS + gmx_MMPBSA. ε_int sensitivity sweep planned. → STRC h01 Phase 9b Holo MD MM-PBSA 2026-04-27 SKELETON
Phase 9c SKELETON. Umbrella sampling PMF along COM-distance reaction coordinate (40 windows × 15 ns), gmx wham analysis. Replaces forced-pull tauRAMD with equilibrium ΔG‡. Path-CV fallback if smoke shows non-radial unbinding. → STRC h01 Phase 9c Umbrella Sampling PMF 2026-04-27 SKELETON
Phase 9d SKELETON. QM/MM single-point + constrained opt on pocket cluster (ωB97X-D/def2-TZVP, electrostatic embedding). LMO-EDA decomposition to test polarization vs charge-transfer share of differential. Disambiguates MM-FF artifacts. → STRC h01 Phase 9d QM-MM Pocket 2026-04-27 SKELETON
Phase 9e SKELETON. Blind neural docking with AF3 + DiffDock + Boltz-2 + Uni-Mol — no seeded coordinates. AF3 + Boltz-2 already integrated; DiffDock + Uni-Mol install pending. AI consensus on K1141-pocket placement is the cross-check signal. → STRC h01 Phase 9e Blind Neural Docking AF3 DiffDock 2026-04-27 SKELETON
Candidate extras menu (not yet skeletoned): MetaD/well-tempered, GaMD, WaterMap/3D-RISM, MSMs from existing MD, LIE, MM-GBSA per-residue decomposition, FTMap/SILCS solvent probes, DCCM/network analysis, force-field cross (AMBER↔CHARMM↔OpenFF), water-model cross (TIP3P↔OPC), charge-isosteric mutational panel (E1659D/K/G), ESMFold/OmegaFold structural-prior comparison. Promote to Phase 9f-9q skeleton if/when load-bearing.

Ranking delta

h01: no change. This is planning + cross-check infrastructure; tier/mech/deliv/misha_fit/next_step do not move until at least one of 9a-9e returns a verdict. Recording the no-change as auditable per AGENTS.md §0.

Artifacts

6 new phase-proof notes under hypotheses/h01-pharmacochaperone/phases/ (1 plan + 5 SKELETON).
No scripts yet (each track adds its scripts to hypotheses/h01-pharmacochaperone/scripts/ per STRC Computational Scripts Inventory discipline when production fires).

2026-04-26 (21:37 — Phase 5q n=20 expansion DELIVERED: σ_within tightened from 1.45× → 1.36× mean / 1.17× median, rank-2 flips adamantyl-Cl > 1-indanyl-Cl, all three τ-tied within SEM)

Phase 5q n=20 overnight DELIVERED. 60/60 UNBOUND in 10.8–59.8 ps biased MD on Phase-5d-truncated 151_e1659a fragment. ~50 min total wall on M5 Max CPU OpenMM (added 45 fresh replicas to the 15 cached from this evening’s pilot, ~1.2 min/rep). 0 censored runs across the trio. Smoke-test (PHASE5Q_N_REPLICAS=6, 1 fresh rep_05 per ligand) passed clean before launch. → STRC h01 Phase 5q v5.3 STRC Within-Target tauRAMD 2026-04-26
σ_within tightened with 4× more samples. Mean-based: 1.45× (n=5) → 1.358× (n=20). Median-based (outlier-robust): 1.167×. Lemma cross-target STRC:TRPM4 bound at 2 × σ_within = 2.72× (mean) or 2.33× (median), well below the >5× kinetic-selectivity gate. v5.3 sulfonamide kinetic-selectivity falsification is now statistically robust at n=20, not n=5-pilot strength. TRPM4 v5.3 dock/τRAMD remains correctly skipped.
τRAMD rank order at n=20: adamantyl_-CF3 (21.59±2.50, median 17.45) > adamantyl_-Cl (17.46±0.60, median 17.20) > 1-indanyl_-Cl (15.90±0.82, median 14.95). Rank-2 flipped from 1-indanyl_-Cl (n=5) to adamantyl_-Cl (n=20) — within-SEM, reflects sampling more of cluster centre. By median, all three are statistically tied (14.95–17.45 ps band). adamantyl_-Cl retains tightest SEM (0.60 ps over 20 reps; geometry-driven kinetic exit reproducibility).
Affinity-rank vs τ-rank split is now load-bearing. Three independent affinity proxies (Boltz-2 dynamic, Vina rigid n=1, Vina ensemble n=20 mut-snapshots) all rank 1-indanyl_-Cl first; τRAMD at n=20 ranks it last by mean and median. The split is no longer a small-n artifact — it is a real method-class divergence between equilibrium-pose-affinity vs kinetic-residence axes. Strengthens the paper-grade “no single winner” lead-committee thesis.
Figure 4 regenerated with n=20 numbers (paper_figure4.{png,svg}). Title/annotations updated to “σ_within = 1.36× (n=20)” with rank order adamantyl_-CF3 > adamantyl_-Cl > 1-indanyl_-Cl.
Phase 5q proof + ranking.md updated to n=20 paper-grade form with median + range columns + rank-flip discussion.
Ranking delta: A held. mech 4 / deliv 3 / misha_fit 4 unchanged. n=20 expansion strengthens existing claims (lemma robustness, lead-committee no-single-winner) but does not move scores. Next P0 (overnight): redeploy quartz wiki with updated Figure 4 + Phase 5q proof so wiki.strc.egor.lol shows n=20 numbers. P1: v5.4 design rationale draft (break cage/linker coupling); α sensitivity analysis on combined-score; blinded retrospective on tafamidis/TTR + lumacaftor/CFTR Coulomb-score predictivity. A_hold_publish_then_v54_design. → STRC h01 Phase 5q v5.3 STRC Within-Target tauRAMD 2026-04-26

2026-04-26 (20:30 — paper-ready batch: Phase 8c v5.3 ADMET + Phase 5e v5.3 mut-ensemble + Figure 4 + paper outline §4.6 + lead committee matrix)

Phase 8c v5.3 ADMET-AI DELIVERED. 3 PASS-list ligands, 10-endpoint gate at 90 percentile drugbank-approved. 2/3 fully clean (0 flags): adamantyl_-Cl, adamantyl_-CF3. Affinity-rank-1 1-indanyl_-Cl has 1 borderline flag (CYP3A4_Veith 90.4 vs gate 90.0; all other 9 endpoints clean). hERG 65–69 percentile across class — no liability separation. AMES separation: 1-indanyl 49.2 vs adamantyl 2–3 (aromatic indane heuristic-positive). → STRC h01 Phase 8c v5.3 ADMET-AI Triage 2026-04-26
Phase 5e v5.3 mut-ensemble re-dock proof DELIVERED. Aggregate JSON existed from Phase 5e_v2 v5_3-top3 (n=20 Phase 5d snapshots, exh=8); proof note formalises the data into paper-grade form. 1-indanyl_-Cl rank-1 by mean (−0.67 ± 3.10) and median (−1.39); adamantyl scaffolds wide-spread (std ≈ 11) — best-pose ≤ −4.6 but high-positive median means snapshot-cherry-picking, not a robust mut-ensemble binder. Three independent affinity proxies (Boltz-2 dynamic, Vina rigid n=1, Vina ensemble n=20) agree on rank-1. → STRC h01 Phase 5e v5.3 Mut Ensemble Dock 2026-04-26
Figure 4 rendered: phase5q_strc_v53/paper_figure4.{png,svg} (2-panel: v5.3 STRC τRAMD per-ligand mean±SEM scatter + lemma cross-target bound 2.9× vs gate 5× with σ_within ≥ 3× threshold overlay). Script: scripts/phase5q_paper_figure4_v53_pre_emptive.py. Reads ranking.json, no new compute.
Paper outline §4.6 added. STRC Paper Draft Outline 2026-04-25 now has v5.3 sulfonamide class delivery section with: lemma second-validation point + lead-committee 2-track tradeoff (affinity-led vs ADMET/kinetic-confidence) + class-level verdict (“kinetic-selectivity rescue strategy at the head-group level is exhausted for adamantyl-/indanyl-cage scaffolds; v5.4 must break geometry coupling at cage/linker, not head”). §2.6 (off-target selectivity) reclassified from “pending” to “closed-by-bound” for v5.2 + v5.3. Outstanding-before-submission list checked off through cross-target lemma row.
Lead committee, paper-grade decision (no single winner): affinity-track = 1-indanyl_acylsulfonamide_SO2Me_-Cl (Boltz-2 + Vina + n=20 ensemble all rank-1; cost = 1 borderline ADMET flag). ADMET/kinetic-confidence track = adamantyl_acylsulfonamide_SO2Me_-Cl (0 flags, tightest τRAMD SEM 0.65 ps; cost = mut-ensemble mean +4.09 means best-pose-only). Both belong on Figure 5 lead summary. Class-level: v5.3 is the first head class that passes both Vina-method-class rule + within-target lemma pre-screen; same as v5.2 on cross-target axis (both bounded ≤ ~3× ≪ 5×).
Ranking delta: A held. mech 4 / deliv 3 / misha_fit 4 unchanged. ADMET axis closes the lead-committee tradeoff structurally: no v5.3 ligand simultaneously affinity-rank-1 + τ-rank-1 + ADMET-clean. This is a scaffold-class limit, not a within-class ranking failure. Next P0: publish to public Quartz wiki (paper-grade html for wiki.strc.egor.lol/h01/); WT-matched ensemble (Phase 5d-WT MD + Phase 5k-WT APBS) for specificity claim closure; α sensitivity analysis. A_hold_publish_then_wt_specificity. → STRC h01 Phase 8c v5.3 ADMET-AI Triage 2026-04-26 + STRC h01 Phase 5e v5.3 Mut Ensemble Dock 2026-04-26

2026-04-26 (20:00 — Phase 5q STRC v5.3 within-target τRAMD: σ_within = 1.45× → cross-target bound ≤ 2.9× ≪ 5×, v5.3 kinetic-selectivity rescue falsified pre-emptively, TRPM4 v5.3 τRAMD SKIPPED)

Phase 5q STRC within-target τRAMD DELIVERED. 3 v5.3 acyl-sulfonamide PASS-list ligands × 5 replicas, all 15 UNBOUND in 7.1–22.3 ps biased MD on Phase-5d-truncated 151_e1659a fragment. ~14 min wall on M5 Max (CPU OpenMM). adamantyl_-CF3 τ̄ 23.88±5.23 (rep-4 outlier 44.5 inflates SEM; sub-mean dropping rep-4 = 18.7), 1-indanyl_-Cl τ̄ 18.00±2.76, adamantyl_-Cl τ̄ 16.46±0.65 (tightest replicate confidence). → STRC h01 Phase 5q v5.3 STRC Within-Target tauRAMD 2026-04-26
Pre-emptive kinetic-selectivity falsification. Within-target spread max/min = 1.45× (with all reps) or 1.14× (drop CF3 rep-4 outlier). Phase 5p activation lemma: cross-target STRC:TRPM4 bounded ≤ 2× σ_within = 2.9×, well below required >5×. Anchor: Phase 5m v5.2 σ_within = 1.7× and observed TRPM4 cross-target ratio = 1.08–1.52× (lemma’s 2× factor was conservative; observed ratio ≤ σ_within). Class falsified by within-target geometry alone. Decision: skip TRPM4 v5.3 dock + τRAMD (~3 h compute saved). Phase 5m v5.2 TRPM4 proof remains the load-bearing kinetic-selectivity verdict for the acyl-sulfonamide line.
Method-class consistency. τRAMD residence rank (CF3 / 1-indanyl-Cl tied at ~18 ps without outlier) does not contradict Phase 5q Boltz-2 + Vina affinity rank-1 (1-indanyl_acylsulfonamide_SO2Me_-Cl). Convergent affinity lead = 1-indanyl_acylsulfonamide_SO2Me_-Cl for next-step lead-committee selection.
Script bug fix. phase5q_strc_taramd_v53.py:537 had invalid f-string {u:.1f if u else '—'} (Python ternary inside format spec); fixed to use a temporary u_str variable. ranking.json was written first (aggregation completed), so data is intact; ranking.md regenerated post-hoc from JSON. Also added missing ~/STRC/models/docking_runs/8e_v5_3/poses/ symlinks → phase5e_v2/v5_3-top3/ligands/*.pdbqt so the script’s DOCK_POSES lookup resolves on next run.
Ranking delta: A held. mech 4 / deliv 3 / misha_fit 4 unchanged. v5.3 sulfonamide class is empirically (now pre-emptively) bounded on the kinetic-selectivity axis; affinity-axis lead is 1-indanyl_acylsulfonamide_SO2Me_-Cl. Next P0: lead-committee selection on 1-indanyl_-Cl — in-silico ADMET (Phase 8c-style ADMET-AI), on-target ensemble docking on Phase 5d mutant snapshots, paper figure update with v5.3 row. NOT TRPM4 v5.3 dock/τRAMD. A_hold_lead_committee. → STRC h01 Phase 5q v5.3 STRC Within-Target tauRAMD 2026-04-26

2026-04-26 (17:xx — h01 state audit + TRPM4 closure reconciliation)

Phase 5m TRPM4 closure reconciled: proof note existed but hub/log still said TRPM4 tauRAMD was launched/pending. Cross-target tauRAMD delivered 2 ligands × 3 replicas on TRPM4 8RD9; tau_TRPM4/tau_STRC = 1.08-1.52x vs required >5x, empirically falsifying v5.2 kinetic-selectivity rescue. Hub next_step updated to v5.3 within-target tauRAMD first, then conditional TRPM4 dock/tauRAMD. A held; mech 4 / deliv 3 / misha_fit 4 unchanged. → STRC h01 Phase 5m TRPM4 Cross-Target tauRAMD 2026-04-26
State audit filed: h01 phase notes normalize cleanly except 12 frontmatter reordering-only changes; 0 missing ranking deltas. Open graph hygiene remains: broken links in h01 hub/log/proofs and stale Dataview snapshots outside h01. No ranking move. → STRC h01 State Audit 2026-04-26

2026-04-26 (13:1x — Phase 5q v5.3 acyl-sulfonamide top-3: Boltz-2 + Vina-v3 rigid AGREE on rank-1 = `1-indanyl_SO2Me_-Cl` ipTM 0.645)

Phase 5q DELIVERED. Boltz-2 (M5 MPS, 23:14 wall, 5 diffusion samples × 3 lig, dimorphite-deprotonated SMILES [N-]) and Vina-v3 rigid-receptor mutant ensemble (20 Phase 5d snap × 3 lig, exh=8, ~38 min wall) on Phase 5p PASS-list top-3 acyl-sulfonamides. All 3 PASS Boltz-2 ipTM ≥ 0.50. Both methods rank v5.3__aq3__1-indanyl__acylsulfonamide_SO2Me__-Cl first by mean (Boltz ipTM 0.645 / pTM 0.855; Vina mean ΔG −0.67, best −4.59). adamantyl_-Cl 2nd by mean on both (Boltz 0.589 σ=0.034 tightest seed-spread; Vina mean +4.09 best −5.60 — best-pose tightest). adamantyl_-CF3 3rd. → STRC h01 Phase 5q v5.3 Acyl Sulfonamide Boltz-2 + Vina Consensus 2026-04-26
Method-class lesson. Vina rigid-receptor AGREES with Boltz-2 on acyl-sulfonamide (anion −1, smaller steric envelope) — opposite of the Phase 5e-v2 v5.2 finding where Vina vs τRAMD anti-correlated on phosphonate (anion −2, bulky tetrahedral head). The rigid-receptor protocol’s failure mode is conditional on head-group magnitude AND volume. Future ZBG-design pre-filter: if anion ≥ −1.5 → Vina usable; if ≤ −2 → defer to Boltz-2 / τRAMD only.
Activation status held. Phase 5p activation gate (Phase 5m STRC:TRPM4 τ-ratio < 5×) remains load-bearing for v5.3 promotion to lead-committee. Phase 5m within-STRC spread 1.7× mathematically bounds cross-target discrimination < 3.4× ≪ 5×, so Phase 5q candidates are biologically-plausible affinity leads but selectivity-rescue is falsified as a class by Phase 5m. Phase 5q gives us which v5.3 candidate to ship to TRPM4 first (1-indanyl_SO2Me_-Cl by mean), not whether to ship the class.
Ranking delta: A held. mech 3 / deliv 4 / misha_fit 4 unchanged. A_hold_pending_TRPM4_taramd. Phase 5q ranks the v5.3 candidates internally; full activation requires TRPM4 τRAMD on lead + top-2 (~5h compute, deferred to next session per CPU-load discipline today — current load 18-23 with 3 Vina runs in flight on D-v3 + B-full-v3, plus τRAMD Metal GPU + Spotlight).
Inventory updated with Phase 5q artifact path and proof-note backreference. Phase 5q candidates now have method-class-justified ranks ready for TRPM4 τRAMD next-compute.

2026-04-26 (10:5x — Phase 5e-v2 v5.2-shortlist: Vina vs τRAMD ρ = −1.000 anti-correlation, Gasteiger zero-anion artifact identified as load-bearing — Vina ΔG axis demoted for anion-leads until Meeko/RESP upgrade)

Phase 5e-v2 generic mutant ensemble re-dock orchestrator landed. New ~/STRC/hypotheses/h01-pharmacochaperone/scripts/phase5e_v2_mutant_redock.py replaces stale ~/Brain/research/strc/models/pharmacochaperone_phase5e_mutant_ensemble_redock.py WORK_DIR. Configurable source: v3b-top50, v5.2, v5.2-shortlist. Outputs to STRC vault. Same K1141 pocket-ring centroid box (18 Å, residues 1141/1135/1137/1165/1167/1175 Cα), same obabel→Gasteiger receptor prep as 2026-04-23 Phase 5e. Smoke at exh=4 → fluke -0.38 kcal/mol on 1 ligand × 1 snap; production at exh=8 cpu=4 produces sane ΔG distributions.
Phase 5m partial-aggregator landed. New phase5m_aggregate_completed.py reads all per_ligand/<lig>/replica_*/result.json rolling, no writes inside per_ligand/ (safe to run while production alive). Confirmed 25/25 UNBOUND, no censored, τ_ps spread 15.4–26.1 mean across 5 v5.2 ligands.
🚨 Phase 5e-v2 v5.2-shortlist (5 lig × 20 snap, finished 02:51 UTC) delivered Vina mut-ensemble ranking. Joined to τRAMD per-ligand stats. Spearman ρ(Vina mean ΔG, −τ median) = −1.000 on n=5 — perfect inversion. Vina top: adamantyl_CONHOH_-Cl ΔG=−4.61 (Kd 417 μM, τ_med 15.7 ps). Vina worst: 4-F-biphenyl_phosphonate_-CN ΔG=+0.30 (Kd 1.6 M, τ_med 20.0 ps). τRAMD reverses this exactly. → STRC h01 Phase 5e-v2 v5.2 Shortlist Vina vs tauRAMD 2026-04-26
Cause: Gasteiger PDBQT total-q ≈ 0.000 ± 0.004 e on anionic phosphonate / acyl-sulfonamide / carboxylate / tetrazolide heads (already a paper-draft caveat — STRC Paper Draft Outline Figure 2). Vina cannot represent the anion → cannot see the +5.99 ± 1.37 kT/e WT→mut electrostatic shift quantified in STRC h01 Phase 5k Ensemble APBS on Phase 5d Mutant MD 2026-04-24 → mutant Vina ranking is electrostatically blind for anion-leads. τRAMD on solvated AMBER14SB+GAFF2 with formal charges does see it → τ-axis ranking is biology-faithful. The two methods don’t disagree on biology, they disagree on what ligand prep tier the chemistry needs.
Implications for D (running): Phase 5e-v2 v3b-top50 (BG PID 14755, ~1.6%, ETA 3-4h) will reproduce this artifact at scale. Most v3b stage2 hits are anionic. Early M result (“11/13 weaker on mutant”, n=1) is the same artifact, not real biology. Treat the v3b-top50 run as methodology smoke for the orchestrator, not a science result. Do NOT write a Phase 5e-v2 v3b verdict on the current Gasteiger run.
Fix path = Phase 5e-v3 with proper formal-charge ligand prep. Priority: (1) Meeko AD4 charges via mk_prepare_ligand.py --add-charge ad4 (pip install meeko into strc-mmgbsa env — not currently installed), (2) Antechamber/RESP for the final 3-6-compound lead committee. After upgrade, expected: Vina rank flips to recover τRAMD ordering on v5.2-shortlist; v3b “11/13 weaker” trend compresses toward 2026-04-23 Phase 5e “max kd_ratio 2.09×, mean 1.49×” pocket-robust verdict.
Ranking delta: A held (mech 3 / deliv 4 / misha_fit 4 unchanged). What changes is interpretive load on each axis: Vina ΔG demoted to “scaffold proxy for neutral fragments only” pending Meeko/RESP upgrade. APBS confirmed as correct electrostatic readout. τRAMD confirmed as load-bearing kinetic axis for anion-leads. Phase 5m within-STRC τ spread 1.7× verdict (rescue falsified) holds and is strengthened by this cross-method convergence. v5.3 acyl-sulfonamide pipeline (Phase 5p) activation status held — still gated on Phase 5m STRC:TRPM4 τ-ratio < 5×. A_hold_ligand_prep_pivot. → STRC h01 Phase 5e-v2 v5.2 Shortlist Vina vs tauRAMD 2026-04-26
Inventory updated: STRC Computational Scripts Inventory now lists phase5e_v2_mutant_redock.py + phase5m_aggregate_completed.py under h01.

2026-04-26 (10:4x — Phase 5m production τRAMD: 25/25 UNBOUND, CONHOH confirmed shortest, within-STRC spread 1.7× falsifies kinetic-selectivity rescue, v5.3 pipeline ACTIVATED)

Production τRAMD on v5.2 mut-prefer + mech-anchor shortlist completed. 5 ligands × 5 replicas, all 25 UNBOUND in 13-55 ps biased MD. Substrate Phase-5d-truncated 151-aa fragment (1066-1216), 14 kcal/(mol·Å) bias, Gasteiger charges, 30-min wall-cap per replica via subprocess. → STRC h01 Phase 5m Production tauRAMD Ranking 2026-04-26
Pipeline architecture lessons (5 SIGSEGV bugs / parser races debugged):
1. AF3-folded substrate → 109k atoms → Apple Silicon Metal-OpenCL SIGSEGV → switched to Phase-5d-truncated 45k-atom substrate (Phase 5l ring-RMSD 0.97 Å between substrates means swap is methodologically clean)
2. am1bcc charges + Precision:mixed → SIGSEGV → fallback to Gasteiger (smoke-validated; relative ranking unaffected; absolute k_off mapping deferred)
3. Precision: "mixed" on equilibration → SIGSEGV → use platform-default precision (smoke pattern)
4. chunk-loop pattern (run_RAMD_sim called per 5000-step chunk) → state re-init, ligand atoms ejected to [-22, 7, -22] nm → switched to single-call run_RAMD_sim (smoke pattern)
5. worker-side log-parse race-conditioned against openmm_ramd’s process-exit log flush → all replicas mis-recorded as censored despite real EXIT events → moved log parsing to parent post-subprocess (canonical source of truth)
Final ranking (τ̄ ± SD ps, n=5):

Rank	Ligand	τ̄ (ps)	ratio vs CONHOH
1	`adamantyl_CONHOMe_-CN`	26.1 ± 16.6	1.70×
2	`4-F-biphenyl_phosphonate_-CN`	21.1 ± 5.4	1.37×
3	`adamantyl_CONHOMe_-Cl` (LEAD)	19.6 ± 9.6	1.27×
4	`1-indanyl_phosphonate_-CF3`	19.0 ± 4.2	1.24×
5	`adamantyl_CONHOH_-Cl` (mech-anchor)	15.4 ± 1.0	1.00×

Three load-bearing findings:
1. CONHOH legacy mech-anchor confirmed shortest residence time. Independent kinetic-axis validation of Phase 7I demotion CONHOH→mech-anchor; lead pivot to CONHOMe-Cl now data-grounded on three orthogonal axes (Boltz-2 ipTM mut-prefer +0.066 + APBS pose-anion-φ −3.24 kT/e improvement on TRPM4 + τRAMD residence-time +27 % over CONHOH).
2. CONHOMe-CN edges out as top τ̄ but with high variance (one 55-ps outlier replica). Robust median 18.7 ps essentially ties with lead median 15.8. Phase 8h-lite #6 had flagged CONHOMe-CN as marginal RWM (P 3.0× TMPA); on the 1 mM IT dose-escalated regime locked in Phase 8h-lite #9, the lower P_RWM is partially compensated. Integrated triage still favours CONHOMe-Cl.
3. Within-STRC τ-spread 1.7× falsifies the kinetic-selectivity-via-existing-head-group rescue hypothesis. With same-class ligands achieving only 1.7× within-target discrimination, cross-target discrimination is mathematically bounded above 5× (typical inter-target factor < 2×; total < 3.4× ≪ 5× target). The Phase 8g-v3-lite τRAMD-rescue hypothesis is now superseded.
v5.3 acyl-sulfonamide pipeline ACTIVATED — Phase 5p PASS-list (6 candidates) graduates from “ready” to “execute”: Boltz-2 + Vina + APBS pipeline on 1-indanyl_acylsulfonamide_SO2Me_-Cl + adamantyl_acylsulfonamide_SO2Me_-Cl + adamantyl_acylsulfonamide_SO2Me_-CF3 is the new heavy P0.
Caveats:
- N=5 replicas is small; SEM ±2-7 ps; only CONHOH (SD 1.0) is unambiguously separable from the other 4
- Off-target τRAMD not run; proper kinetic-selectivity test requires lead × TRPM4 / TMEM16A. Queued P0-light
- Gasteiger charges; absolute k_off mapping requires am1bcc; relative ranking should hold

Ranking delta

tier A → A (held)
mech 4 → 4 (held; multi-axis lead pivot data-grounded)
deliv 3 → 3 (held)
misha_fit 4 → 4 (held)
next_step refresh: drop “τRAMD on v5.2 shortlist” (CLOSED — within-STRC 1.7× falsifies same-head kinetic rescue); add “Phase 5q: v5.3 acyl-sulfonamide Boltz-2 + Vina + APBS on 6 PASS-list candidates” P0; add “TRPM4 τRAMD on lead + top-2 alternatives” P0-light (kinetic-selectivity number for paper); add “NAGL .pt model for am1bcc-grade charges” P1-light infrastructure.

Artifacts

artifacts/phase5m_production/per_ligand/<lig>/replica_<k>/{ramd.log, trajectory.dcd, result.json} (25 replicas)
artifacts/phase5m_production/ranking.{json,md}

2026-04-26 (09:4x — parallel session while τRAMD production runs: calnexin retention NEGATIVE, v5.3 acyl-sulfonamide library 6/12 PASS, sci-hub 4 ERAD-timing PDFs landed)

Three parallel deliverables while Phase 5m production τRAMD compute is running (5 ligands × 5 replicas on Phase-5d-truncated 151-aa K1141 fragment; cache-cleared restart 09:47 after AF3-folded substrate hit 109k-atom OpenCL SIGSEGV — Phase 5l ring RMSD 0.97 Å says local pocket geometry is identical so substrate swap is methodologically clean).

(a) Phase 5n-lite — N-glyc map + calnexin retention proxy on STRC

Script: scripts/phase5n_lite_nglyc_calnexin_map.py. Maps NX[ST] (X≠P) sequons on 1775-aa STRC, distances from each to E1659 + K1141 in 3D from Phase 5d snap_010, WT vs E1659A diff. → STRC h01 Phase 5n-lite Calnexin Retention NEGATIVE 2026-04-26
Result: NEGATIVE. STRC has 14 sequons; 0 in C-terminal ZP-like domain (1501-1771) where E1659A lives; nearest sequon (N1179, NLT motif) is 29.8 Å from E1659 — outside the ~25 Å canonical calnexin engagement radius. E1659A creates 0 / destroys 0 sequons — does not perturb N-glyc landscape.
Implication: calnexin extended-retention is NOT the rate-limiting step for E1659A QC clearance. The mannose-trimming-clock mechanism that gates calnexin → EDEM → HRD1 progression cannot operate on a domain that calnexin doesn’t see.
Re-ranking of QC rate-limiting-step candidates from STRC E1659A QC Rate-Limiting Step Lit Gap-Map: ↓ calnexin extended-retention (was top); ↑ HRD1/SEL1L core ERAD; ↑ ER-phagy (FAM134B/RTN3 selective autophagy); ↑ BiP/Grp78 retention (sequon-independent recognises exposed hydrophobic patches on misfolded ZP-like fold).
τ-floor 60 min holds (per Ward & Kopito 1994 F508del CFTR ERAD t½ 30 min × 2× safety) but the mechanism gating that floor pivots from calnexin-cycle to BiP-cycle / HRD1-recognition. Lit basis updates to Hebert et al 1996 + Christianson 2008 + Schmidt 2024 BiP review.
Pharmacochaperone binding at K1141 even safer than Migalastat Dosing-Cycle Principle flagged — K1141 is in central FN3-like with 4 nearby sequons; not on the ZP-like fold disordered by E1659A. Allosteric K1141 binding cannot interfere with the (now BiP- or HRD1-driven) ER-residence-time mechanism on the ZP-like domain.

(b) Phase 5p — v5.3 acyl-sulfonamide library design + triage

Script: scripts/phase5p_v53_weakanion_library.py. RDKit-only library construction + Lipinski/TPSA/anion-count gates. → STRC h01 Phase 5p v53 Acyl Sulfonamide Library 2026-04-26
Strategy: dilute the head-group anion charge density across more atoms so highly-cationic off-target pocket interiors (TRPM4, TMEM16A) pull the head-group less. Acyl sulfonamide is canonical hydroxamate bioisostere (pKa 4-5; charge delocalised across C=O + 2× S=O + N — 4 atoms vs CONHO⁻ single-O).
Library: 2 heads (acylsulfonamide-SO2Me, -SO2CF3) × 2 bodies (adamantyl, 1-indanyl) × 3 tails (-Cl, -CN, -CF3) = 12 SMILES. Ring-bioisostere heads (tetrazole, oxadiazolone, hydroxytriazole) require ring-numbering-aware SMILES assembly that the simple prefix-replacement approach cannot do reliably; deferred to Phase 5p-v2 with hand-written SMILES if required.
Pass rate 6/12 = 50 %. All -CN tails fail TPSA (TPSA 111.4 > 100 cap). Two -CF3 + SO2CF3 combos fail Lipinski MW (501-521 > 500 cap).
Top-3 next-gen lead candidates (ranked by RWM-permeability balance):
1. v5.3__aq3__1-indanyl__acylsulfonamide_SO2Me__-Cl — MW 415.9, logP 1.48, TPSA 87.6
2. v5.3__aq3__adamantyl__acylsulfonamide_SO2Me__-Cl — MW 434.0, logP 1.51, TPSA 87.6 (direct head-swap of the v5.2 lead)
3. v5.3__aq3__adamantyl__acylsulfonamide_SO2Me__-CF3 — MW 467.5, logP 1.88, TPSA 87.6
Activation gate: Boltz-2 + Vina + APBS pipeline on top-3 only triggers if Phase 5m τRAMD STRC : TRPM4 ratio < 5× on v5.2 shortlist. Otherwise stay on CONHOMe-Cl lead.

(c) Sci-hub ERAD-timing literature — 4 PDFs

4 papers retrieved via the sci-hub skill (sci-hub.ru mirror, magic-byte validated %PDF):
- Ward & Kopito 1994 — ~/BookLibrary/incoming/ward1994.pdf (3.85 MB). DOI corrected to 10.1016/S0021-9258(18)47306-1 (47307-7 was wrong in lit-gap-map). F508del CFTR ERAD t½ ≈ 30 min — canonical τ-floor calibration source.
- Lukacs et al 1994 — lukacs1994.pdf (2.61 MB). CFTR ER conformational maturation kinetics; companion to Ward & Kopito.
- Hebert, Foellmer & Helenius 1996 — hebert1996.pdf (1.96 MB). Calnexin cycle on HA microsomes; canonical CNX/CRT mechanism source. Re-purposed: was needed for calnexin-axis interpretation, now informs the BiP-axis pivot (Hebert later work covers BiP-cycle handoff).
- Molinari 2007 — molinari2007.pdf (667 KB). Glycan-driven calnexin retention timing review. Now lower priority post-Phase-5n-lite NEGATIVE.
BookLibrary daemon will pick up for MinerU → RAG → Brain stub; downstream auto Brain extraction.
Lit-gap-map note [[STRC E1659A QC Rate-Limiting Step Lit Gap-Map]] to be updated post-MinerU with Ward & Kopito τ-floor and Hebert BiP handoff details, as ranking-delta footnotes.

Production status (in progress)

Phase 5m production τRAMD: 5 ligands × 5 replicas. Substrate Phase-5d-truncated 151-aa fragment (smoke-validated 45k atoms; AF3-folded substrate failed at 109k atoms / OpenCL SIGSEGV on Apple Silicon Metal). Phase 5l ring RMSD 0.97 Å between AF3 and Phase-5d substrates means local pocket geometry is identical — substrate swap clean.
Restart 09:47 (PID 90510). Lead adamantyl-CONHOMe-Cl building system + am1bcc charging now (now functional via ambertools conda install + PATH fix in script). ETA 4-6 hr full run.

Ranking delta

tier A → A (held)
mech 4 → 4 (held; Phase 5n-lite NEGATIVE confirms E1659A is fold-defect not glyc-defect, supporting the thermodynamic-pocket framing per Phase 5k matched-ensemble +5.99 kT/e)
deliv 3 → 3 (held)
misha_fit 4 → 4 (held)
next_step refresh: drop “Phase 5n-lite calnexin/CRT MSA scan” (CLOSED NEGATIVE); add “BiP-cycle / HRD1-SEL1L lit retrieval (Christianson 2008, Schmidt 2024)” P1-light; v5.3 acyl-sulfonamide PASS-list (6 SMILES) stage-1 ready, activation gated on Phase 5m STRC:TRPM4 τ-ratio.

Artifacts

artifacts/phase5n_lite/result_nglyc_map.{json,md}
artifacts/phase5p/v53_library.{csv,json,md}
~/BookLibrary/incoming/{ward1994,lukacs1994,hebert1996,molinari2007}.pdf
Phase 5m production running (no artifact yet)

2026-04-26 (09:0x — Phase 5l + 5m smoke: AF3 fragment ladder selected, τRAMD pipeline validated, lead unbinds at 17.9 ps biased on 151-aa fragment)

Closing the residence-time gap surfaced as P0 by Phase 8g-v3-lite (kinetic-selectivity escape hatch). Two interlocked deliverables. → STRC h01 Phase 5l + 5m Smoke Fragment AF3 + tauRAMD Pipeline 2026-04-26

(a) Phase 5l — AF3 K1141 fragment ladder + selection gate

8 AlphaFold Server v3 jobs (4 fragment lengths × WT/E1659A) submitted via ~/.claude/skills/alphafold-server/SKILL.md (Playwright MCP). All 8 finished in 2 min wall (small single-domain proteins). 40 CIFs + 40 confidence JSONs to ~/STRC/models/af3_jobs_2026-04-26_phase5l/downloads/. Submission log at same path.
Selection gate scripts/phase5l_selection_gate.py (gemmi CIF parser): pocket pLDDT ≥ 80, K1141-ring Cα RMSD < 1.5 Å, whole-fragment Cα RMSD < 2.5 Å, all vs Phase 5d snap_010 full-length E1659A.
Strict-gate result: NO fragment passes (whole-frag-RMSD floor 4.87 Å). AF3 folds single-domain fragments without cross-domain context (FN3-like / ZP-like architecture), so global Cα diverges. But K1141-ring local geometry IS preserved: 151_e1659a ring RMSD 0.97 Å; 221_e1659a 1.40 Å; 121_e1659a 3.35 Å (too small).
Relaxed operational gate (ring RMSD < 1.5 Å + pocket pLDDT ≥ 78, drop strict whole-frag gate as it’s not load-bearing for ligand binding): 151_e1659a wins (ring RMSD 0.97 Å, pocket pLDDT 78.7, pTM 0.65).
WT vs E1659A pLDDT differs by ≤ 1 across the ladder; AF3 does not see E1659A as a fold-disrupting mutation at fragment level (consistent with Phase 5d K1141 NZ SD 1.1 Å across 20 snapshots — packing-defect, not fold-disrupting mutation).
Phase 5l artifacts: artifacts/phase5l/selection_results.{json,md}.

(b) Phase 5m smoke — τRAMD pipeline end-to-end validation

Goal: validate pipeline on real fragment + lead pose before committing v5.2 shortlist run (10 lig × 10 rep × 5 ns ≈ 500 ns total).
Substrate: Phase 5d snap_010 hand-truncated to chain-A 1066-1216 (151 aa, K1141@76); PDBFixer added missing atoms + termini + H@pH 7.4. Phase-5d-truncated rather than AF3-folded for the smoke (Phase 5l ran in parallel; truncation trivially preserves Phase-5d MD-relaxed pocket geometry).
Ligand: SMILES from phase8d_v5_2_library.csv row 31, openff Molecule.from_smiles + 1 conformer, COM translated to Phase 8e Vina pose centroid [16.41, -45.38, 10.79] Å, Gasteiger charges (am1bcc requires AmberTools/sqm not installed; same fallback as Phase 5d/5e/8e).
System: amber14SB + tip3p + 0.15 M NaCl, HMR=3 amu, 2 fs timestep, 45 714 atoms total, OpenCL platform.
Equilibration 100 ps NVT @ 310 K; production τRAMD via openmm_ramd.openmm_ramd.RAMDSimulation (canonical at github.com/seekrcentral/openmm_ramd; tool card [[RAMD-OpenMM]] updated — original Kokh 2018 supplement URL HITS-MCM/ramd_openmm is dead). Force 14 kcal/(mol·Å), rMinRamd 0.025 nm, ramdSteps 50, maxDist 4.0 nm.
Result — UNBOUND at MD step 8950 = 17.9 ps biased MD. RAMD log shows lead COM drift [1.279, -4.585, 0.501] nm → [2.293, -4.111, 4.377] nm (4.07 nm displacement) → exit. Wall ~10 min total (build + 100 ps equil + 17.9 ps biased).
One replica is not statistically meaningful — single τ_obs = 17.9 ps establishes ballpark + validates pipeline. Production needs 10 replicas/ligand for SEM.

Pipeline lessons captured

am1bcc charging fails without AmberTools — Gasteiger fallback OK for ranking, install ambertools for paper-level final
PDBFixer required for hand-truncated fragments (raw chain-cut PDB lacks proper N/C-cap templates)
charge_from_molecules must go via template_generator_kwargs in SystemGenerator
RAMDSimulation requires properties dict with platform ({"Precision": "mixed"} for OpenCL)
sim.RAMD_step is method not attribute — extract exit step from smoke_ramd.log regex
Apple Silicon Metal-OpenCL: ~50 RAMD-evaluations/min (≈ 5 ps biased MD/min) on 45 k-atom system
ramd-openmm canonical install: pip install git+https://github.com/seekrcentral/openmm_ramd.git

Ranking delta

tier A → A (held)
mech 4 → 4 (held; AF3 fragment ladder confirms K1141 pocket geometry preserved across single-domain fold)
deliv 3 → 3 (held; selectivity gate FAIL still binding constraint)
misha_fit 4 → 4 (held)
next_step refresh: “Phase 5m production (10 replicas × 14 v5.2 shortlist + diflunisal pos-ctrl + tafamidis-analog neg-ctrl on 151_e1659a AF3 substrate; install AmberTools for am1bcc; cross-substrate sensitivity (1 rep × 4 lig on Phase-5d-truncated for substrate-independence check); est. 5-25 hr depending on parallelism); Boltz-2 binding-mode at TRPM4 + TMEM16A queued P0-light”

Artifacts

artifacts/phase5l/selection_results.{json,md}
artifacts/phase5m_smoke/{smoke_fragment.pdb, smoke_ligand.sdf, smoke_system.xml, smoke_trajectory.dcd, smoke_ramd.log, smoke_log.csv, smoke_result.json}
models/af3_jobs_2026-04-26_phase5l/downloads/* (40 CIFs + 40 confidence JSONs)
models/af3_jobs_2026-04-26_phase5l/submission_log.json (8 AFS job IDs + URLs + local files)
Tool card [[RAMD-OpenMM]] repo URL fixed (HITS-MCM dead → seekrcentral)

2026-04-26 (08:3x — Phase 8g-v3-lite + 8h-lite #9: gate diagnostic + dose-escalation rescue. Heavy-queue replanned)

Two-script session closing the off-target-selectivity diagnostic and the PBPK exposure question. Wall-time ~3 min compute + write-up. → STRC h01 Phase 8g-v3-lite + 8h-lite 9 Pocket Max Phi + Dose Sweep 2026-04-26

(a) Phase 8g-v3-lite — pocket-max-φ resample

Script: scripts/phase8g_v3_pocket_max_phi.py. Reuses existing Phase 8g-v2 (CONHOH) and Phase 7i (CONHOMe) APBS potential.dx grids — no new APBS compute, only resampling + solvent-accessibility mask (1.5 Å ≤ d_to_protein ≤ 4.0 Å; 50–64 % of voxels accessible).
Diagnostic question this answers: is Phase 8g-v2 FAIL pose-driven (Vina put anion-O in sub-optimal voxel; site-defined re-dock would close gate) or residue-driven (off-target pocket interior intrinsically more positive; head must change)?
Result: residue-driven on both variants. CONHOH STRC top-10% +29.95 vs TMEM16A +35.19 → margin −5.24 kT/e. CONHOMe STRC top-10% +21.90 vs TRPM4 +24.39 → margin −2.49 kT/e. Worst-off-target swaps with head group (CONHOH→TMEM16A, CONHOMe→TRPM4).
STRC K1141 pocket is electropositive (+22 / +30 kT/e top-10%; Phase 5k matched-ensemble +5.99 ± 1.37 was the centre-mean — top-10% naturally higher). Mech-4 holds. But pocket ceiling (+22) ≤ TRPM4 ceiling (+24) → site-defined re-dock cannot flip verdict, only narrow margin.
Phase 8g-v3 full re-dock STOPPED. Capital re-allocated. The hub frontmatter next_step claim “Phase 8g-v3 site-defined re-dock + Phase 8e-v2 pose-anion re-rank” is superseded.
Kinetic-selectivity escape hatch elevated to load-bearing. Per Pharmacochaperone Residence Time Criterion, slow STRC k_off + fast off-target k_off can deliver clinical selectivity without thermodynamic affinity selectivity. τRAMD on v5.2 shortlist promoted P1-light → P0.

(b) Phase 8h-lite #9 — PBPK dose-escalation sweep

Script: scripts/phase8h_lite/cochlear_pbpk_dose_sweep.py. Re-uses identical 8-compartment PBPK topology + rate constants from Phase 8h-lite #7. Sweeps applied IT dose × [100 µM, 1 mM, 10 mM] × [fast / default / slow apical-uptake bracket].
Two ceilings tracked: HC_ER ≥ 1 µM (therapeutic-window fraction, target compartment) and HC_CYT ≥ 50 µM (safety flag from Givinostat mARC1 2 Hydroxamate Bypass: 10× margin on 4 µM hydroxamate Cmax MTD).
IT dose precedent: dexamethasone IT 10–60 mM (OTO-104), SPI-1005 ebselen IT up to 30 mM. 1 mM and 10 mM applied are well inside IT pharmacy.
Result — therapeutic-window matrix (% of 72 h with HC_ER ≥ 1 µM):

Apical scenario	100 µM	1 mM	10 mM
fast (t½ 30 min)	38.9 %	96.9 %	98.5 %
default (t½ 95 min)	0.0 %	95.8 %	98.0 %
slow (t½ 300 min)	0.2 %	93.8 %	97.4 %

Safety flag (% of 72 h with HC_CYT ≥ 50 µM): 0 % at 100 µM and 1 mM in all apical brackets. 65.5 / 93.7 % at 10 mM in default / fast — 10 mM is the safety cliff. Only slow-apical bracket safe at 10 mM.
1 mM IT dose is the sweet spot — closes window across all apical brackets without breaching cytosol safety. Phase 8h-lite #7 deliv-3 ceiling on the exposure axis is liftable purely through dose escalation; no new chemistry / permeation enhancer required.
1 mM is locked as the reference dose for all subsequent PBPK / paper claims (supersedes the Phase 8h-lite #7 100 µM reference).
But deliv axis cannot lift to 4 until selectivity gate (8g-v3-lite FAIL) closes — even 1 mM exposure is meaningless if TRPM4 anti-selective electrostatics translate to actual binding.

Ranking delta

tier A → A (held)
mech 4 → 4 (held; STRC pocket electropositivity confirmed at residue-driven level)
deliv 3 → 3 (held; exposure rescued at 1 mM IT but selectivity gate FAIL prevents axis lift)
misha_fit 4 → 4 (held)
next_step replaced: “τRAMD v5.2 shortlist (P0, kinetic-selectivity load-bearing) + Boltz-2 binding-mode at TRPM4 (CONHOMe lead) and TMEM16A (CONHOH lead) for top-2 Tier-1 (P0-light, does electrostatic ceiling translate to ipTM binding?) + 1 mM IT dose locked as reference + v5.3 weakly-anionic head-group exploration (acyl sulfonamide, oxadiazolone) (P1-light) + DEFERRED: Phase 8g-v3 full re-dock (foregone-conclusion compute)“

Artifacts

artifacts/phase8g_v3/pocket_max_phi.json and pocket_max_phi.md
artifacts/phase8h_lite/result_09_pbpk_dose_sweep.json and result_09_pbpk_dose_sweep.md

2026-04-26 (01:5x — DR1 next_step delivered: PBPK ODE + interface map + lit-gap-map. deliv-3 ceiling reinforced from delivery side)

Three DR1-surfaced next_step tasks executed in single session. All P1-light, all on local Mac silicon, no cloud. Cumulative wall-time ~5 min compute + 20 min writeup.

(a) Phase 8h-lite #7 — Cochlear-compartment PBPK ODE

Script: scripts/phase8h_lite/cochlear_pbpk_ode.py (Python + scipy.integrate.solve_ivp). 8-compartment topology: DEPOT (P407 gel reservoir 300 µL) → MEC (middle-ear cleft free pool 100 µL) → PER_B ↔ PER_A (perilymph 35/35 µL) ↔ ENDO (scala media 10 µL) → HC_CYT ↔ HC_ER (10⁻⁵ / 10⁻⁶ mL) → BLOOD sink. Reversible flows where physically required (RWM forward/reverse, ER ↔ cytosol with Kp = 1).
Result: at standard 100 µM applied IT dose, HC_ER Cmax = 0.4–1.3 µM across apical-permeability bracket (t½ 30–300 min); default scenario = 0 % time-above-1-µM-EC over 72 h, fast-apical = 39 %. Bottleneck identified: eustachian-tube clearance from MEC (k = 1.16e-2 /min) is ≈ 11 000× faster than RWM crossing (k = 1.0e-6 /min) — free drug in the middle-ear cleft is leaving via the eustachian tube far faster than it diffuses into perilymph. PER_A (apical turn, speech frequencies) effectively unreached: Cmax 3×10⁻⁵ µM. Mass balance 100.00 % verifies ODE correctness.
Implications: Phase 8h-lite #2 RWM permeability number was load-bearing only on the boundary itself — once cascaded through the full cochlear topology it does not predict cochlear exposure. deliv-3 ceiling reinforced from the delivery side independently of Phase 8g-v2 selectivity FAIL. Three improvement levers identified, each P1-light:
- (i) 10× applied dose (100 µM → 1 mM, well within IT injection literature ranges from steroid precedents)
- (ii) 3× P_RWM via permeation enhancer (0.5% sodium caprate, DMSO co-solvent)
- (iii) sub-µM Kd via v5.3 (CONHOMe / phosphonate Tier-1)
- Combined → 30–100× window widening; lifts default-apical fraction-time from 0 % → ~95 %.
Output: artifacts/phase8h_lite/result_07_cochlear_pbpk.md + .json. ELX-02 lesson officially landed.

(b) Phase 8h-lite #8 — K1141 vs partner-interface map

Script: scripts/phase8h_lite/k1141_interface_map.py (manual PDB column parser; freesasa not installed → quartile-based heavy-atom-burial proxy for SASA ranking). Loaded models/artifacts/phase5d_snapshots/snap_010.pdb (1775-residue full-length E1659A, 27 299 atoms; verified K1141 NZ + A1659 CA present at coordinates).
K1141 NZ → A1659 CA distance: 13.91 Å. Pocket-shell inventory: 8 res ≤ 6 Å, 18 res ≤ 8 Å, 47 res ≤ 12 Å.
Domain assignments (UniProt Q7RTU9 SMART/Pfam): K1141 → central FN3-like (≈ 800–1200); E1659A → C-terminal ZP-like (≈ 1500–1771). They are in different domains, ~14 Å apart in 3D — the lead does not bind to the mutation site, it binds to a neighbouring-domain electropositive pocket whose physics is altered by the mutation’s destabilising effect on the broader fold.
K1141 SASA-proxy = 120 (partially buried), sits in middle quartile of pocket-shell SASA distribution. Halgren druggable-pocket signature — enclosed-yet-reachable. Surface-exposed pocket-shell runs in FN3-like = max 3 contiguous residues (1125–1128). No run ≥ 5 contiguous residues = no canonical partner-binding interface patch at the K1141 region. Jones & Thornton 1996 / Lo Conte 1999 5-residue rule.
CFTR F508del + Elexacaftor architectural analog established: corrector pocket ≠ mutation site, distinct domains, allosteric small-molecule rescue. STRC E1659A + K1141 binder sits in the same logic. Direct paper-§3 citation enabled.
Migalastat Dosing-Cycle Principle design-constraint check passes: K1141 pocket is enclosed within FN3-like domain core, in a different domain from the canonical ZP-like partner-binding region (where E1659 sits). A chaperone bound at K1141 is unlikely to disrupt tectorial-membrane assembly post-trafficking. Slow k_off shifts from double-edge to pure asset → Tafamidis-style design space, not Migalastat constrained-dissociation space. Big design-side simplification.
Output: artifacts/phase8h_lite/result_08_k1141_interface_map.md + .json.

(c) STRC E1659A QC rate-limiting step — lit-mining gap-map

Per-vault grep confirms: no primary literature on stereocilin ERAD timing in ~/STRC/sources/lit/ as of 2026-04-26. Gap is real.
Created STRC E1659A QC Rate-Limiting Step Lit Gap-Map note in ~/STRC/notes/. Structured retrieval queue with canonical DOIs covering 5 areas: (A) calnexin/calreticulin cycle (Hebert/Helenius 1995, Molinari/Helenius 1999, Caramelo/Parodi 2003, Hebert/Molinari 2007 review); (B) ERAD substrate selection (Olzmann/Kopito 2013 review, Hosokawa 2001 EDEM1, Christianson 2008 Hrd1/Sel1L, Bhamidipati 2005 OS-9); (C) F508del CFTR proxy timing (Ward/Kopito 1994 t½ 30 min, Lukacs 1994 EMBO J, Pind 1994, Sun 2008); (D) STRC-specific (Verpy 2001/2008, Avan 2019 mouse, Kammerer 2012 CEACAM16, Lukashkin 2012); (E) ZP-like domain folding (Bork/Sander 1992, Jovine 2002, Han 2010).
Operative interpretation pending retrieval: F508del CFTR ERAD t½ ≈ 30 min is the working τ-floor proxy → τRAMD residence-time floor = 60 min (2× the proxy). Stereocilin is predicted to be a calnexin substrate by domain-architecture analogy (multi-N-glyc + ZP-like + secreted ECM glycoprotein class); direct measurement requires anti-calnexin co-IP from iPSC otic-organoid lysate (wet-lab gate, not compute).
Best-current rate-limiting-step hypothesis: extended calnexin retention → EDEM-mediated mannose-trimming clock → Hrd1/Sel1L recognition. K1141 binding likely intercepts the mannose-trimming clock step by lowering mutant fold ΔG enough that calnexin-cycle re-entry occurs before EDEM fires. Hypothesis, not measurement.
Output: vault note + named queue items for sci-hub / annas-archive skill retrieval. Phase 5n-lite (calnexin/CRT MSA scan on stereocilin sequence) flagged as ~1 hr literature/sequence task.

Summary numerics

Task	Output type	Compute time	Key number
(a) PBPK ODE	new compute + new note	<2 s	HC_ER Cmax 0.4–1.3 µM at 100 µM IT dose
(b) Interface map	new compute + new note	<2 s	K1141–A1659 distance 13.91 Å, different domains
(c) Lit gap-map	structured retrieval queue	manual	F508del τ½ 30 min → h01 τ-floor 60 min

Ranking delta

tier: A (held)
mech: 4 (held — Phase 8h-lite #8 confirms the K1141 pocket is structurally allosteric to the canonical ZP-like partner interface; this strengthens the mech-4 framing for the Tafamidis-style argument once rate-limiting step is named, but does not move the needle to mech 5 without the literature retrieval closure on calnexin/EDEM substrate status. The “different-domain, allosteric pocket” claim is a structural-evidence asset for paper §3.)
deliv: 3 (held — Phase 8h-lite #7 is an independent reason for the deliv-3 ceiling working from the delivery side. Phase 8g-v2 was selectivity FAIL; Phase 8h-lite #7 is exposure FAIL at standard dose. Both must be addressed before deliv 4 returns. Three concrete levers identified.)
misha_fit: 4 (held)
next_step refresh:
- P0 (heavy, blocker): Phase 8g-v3 site-defined re-dock + Phase 8e-v2 STRC pose-anion-O re-rank — closes the off-target FAIL margin (unchanged).
- P0-light (added by Phase 8h-lite #7): dose-escalation sweep on PBPK — re-run cochlear ODE at 1 mM and 10 mM applied IT dose; verify HC_ER stays linear and confirm fraction-time-above-EC reaches > 50% in default-apical scenario.
- P0-light (added by Phase 8h-lite #7): Sensitivity sweep on apical permeability — bracket via in-vitro HEI-OC1 / UB-OC1 cellular uptake assay timeline (wet-lab gate, captured in STRC h01 Phase 8 Wet-Lab Triage SOP).
- P1-light (added by lit-gap-map): retrieve via sci-hub skill the F508del CFTR timing primary literature (Ward & Kopito 1994 first; Lukacs 1994 second). Ingest as ~/STRC/sources/lit/ entries per Lit Audit Status 2026-04-25 schema.
- P1-light (added by lit-gap-map): Phase 5n-lite — calnexin/CRT consensus MSA scan on stereocilin sequence (~1 hr lit + sequence task).
- Existing queue items hold: Phase 5d ensemble re-dock on Tier-1 (CONHOMe / phosphonate); ADMET re-check on CONHOMe class; v5.3 design freeze conditional.
Tools & inventory: 2 new scripts logged in STRC Computational Scripts Inventory § h01 (cochlear_pbpk_ode.py, k1141_interface_map.py). 2 new result notes (result_07_cochlear_pbpk.md, result_08_k1141_interface_map.md) and 1 new atomic note (STRC E1659A QC Rate-Limiting Step Lit Gap-Map) added to vault. No new tool cards (scipy, biopython already documented as standard Python stack).

2026-04-26 (01:3x — DR1 distillation: 5 clinical-precedent atomic notes; rate-limit framing + PBPK gap surfaced)

DR1 read in full (artifacts/deep-research/DR1_pharmacochaperone_precedents.md, 212 lines, 10 clinical-stage pharmacochaperone post-mortems + 5 cross-cutting themes). Per feedback_strc_no_wetlab no vendor/cost content imported — extracted five durable scientific principles directly applicable to H01.
5 atomic notes created in ~/STRC/notes/:
- Tafamidis Kinetic-Stabilization Paradigm — rate-limiting-step framing. Tafamidis works by raising kinetic barrier of TTR tetramer dissociation (the rate-limit). H01 mech-4 is thermodynamic (Phase 5k anion preference); does not name which step on E1659A QC pathway the K1141 binding intercepts. Paper §3 must name the rate-limiting step or accept Migalastat-style empirical framing.
- Migalastat Dosing-Cycle Principle — residence-time ceiling complement to floor in Pharmacochaperone Residence Time Criterion. Migalastat: ER pH 7.4 → lysosomal pH 4.0 drives chaperone dissociation at site of action. Afegostat failure post-mortem: nM-affinity that didn’t dissociate at lysosomal pH became permanent inhibitor. STRC is extracellular — no pH gradient available. H01 must rely on tuned intrinsic k_off OR allosteric binding outside stereocilin-stereocilin interfaces. Choice not yet made.
- Givinostat mARC1 2 Hydroxamate Bypass — pediatric-chronic regulatory precedent for CONHOH. Givinostat (Duvyzat, FDA Mar 2024) cleared via mARC1/2 reduction → amide → carboxylesterase hydrolysis, bypassing CYP450 / Lossen rearrangement / hepatotoxicity. H01’s CONHOH lead inherits this script; in-vitro mARC1/2 assay is the IND-package gate. Companion to Hydroxamic Acid Divalent Cation Liability (covers Zn²⁺/Ca²⁺ off-target axis; this note covers metabolic-safety axis).
- Ataluren Reporter-Stabilization Trap — wet-lab decision-gate hardening. Ataluren bound the firefly luciferase reporter directly and produced a false-positive signal that survived to Phase 3 (FDA rejected; EMA conditional approval non-renewed 2024-2025). Mandates reporter-free native readout (anti-stereocilin Western / mass-spec PRM on iPSC otic-organoid lysates) for every H01 wet-lab gate; refines STRC h01 Phase 8 Wet-Lab Triage SOP. In-silico analog connects to Phase 7I Vina WT-bias artefact: every binding claim from a single scoring function must pass orthologous re-evaluation.
- ELX-02 In Vitro Clinical PBPK Disconnect — in-vitro→clinic gap is owned by tissue partitioning, not affinity. ELX-02 had perfect CFTR-G542X organoid rescue and failed Phase 2 CF because aminoglycoside-class polarity drives renal partitioning, achieving only ~20% of EC₅₀ in pulmonary epithelium. H01 has RWM permeability (Phase 8h-lite §2) but no cochlear-compartment PBPK model — basal→apical perilymph diffusion + endolymph crossover + hair-cell-ER deposition unmodeled. Closed-form ODE PBPK in Python is a few-hours job, no new wet-lab data needed.
Hub ## evidence updated with five new bullets pointing at the notes. Hub next-step not edited in frontmatter — current text already covers Phase 8g-v3 / Phase 5d-on-Tier1 (the FAIL margin from Phase 8g-v2 is the higher-priority active blocker).
Cross-link to existing DR notes: Pharmacochaperone Residence Time Criterion (DR3) and Migalastat Dosing-Cycle Principle (DR1) jointly bracket the k_off design window — floor ≈ ER transit (tens of min – h), ceiling ≈ tectorial-membrane assembly (h – d) for extracellular STRC. τRAMD on the v5.2 / v5.3 shortlist gains sharper interpretation as a result.

Ranking delta

tier: A (held)
mech: 4 (held — DR1 reframes the gap as “rate-limiting QC step not named” and “k_off ceiling not bounded for extracellular target”. Both are unquantified mechanistic axes; neither weakens the Phase 5k anion-preference proof, neither yet strengthens it to mech 5.)
deliv: 3 (held — already downgraded by Phase 8g-v2 APBS off-target FAIL. ELX-02 PBPK lesson reinforces the deliv-3 ceiling: RWM crossing alone is insufficient; cochlear-compartment partitioning unmodeled.)
misha_fit: 4 (held)
next_step refresh — three new compute / lit additions surfaced by DR1, all P1-light unless noted:
- (a) Cochlear-compartment PBPK ODE model in Python (multi-compartment: middle-ear depot → perilymph basal/apical → endolymph → blood via stria-vascularis capillary). Parameterize from RWM permeability (Phase 8h-lite §2) + literature blood-perilymph barrier rates. Output: predicted free-drug C(t) at hair-cell ER vs. lead in-silico ΔG-implied EC. Few-hours job, no new wet-lab data. P1-light, run after Phase 8g-v3 unblocks.
- (b) Stereocilin-partner interface map. Project all known stereocilin-stereocilin / stereocilin-CEACAM16 / stereocilin-OTOG / stereocilin-OTOGL contact patches onto the AF3 E1659A model. Test whether K1141 pocket overlaps any. Determines whether Migalastat-style allosteric escape is available for H01. ~1 hr, leverages existing AF3 + published interactome. P1-light.
- (c) E1659A QC rate-limiting step assignment. Lit-mining first: calnexin/calreticulin/EDEM cycle timing for similar ECM-class missense mutants. Names the step Tafamidis-style mech argument requires. Wet-lab pulse-chase is in STRC h01 Phase 8 Wet-Lab Triage SOP gate but not yet planned with this framing.
Existing queue items hold positions: Phase 8g-v3 site-defined re-dock + Phase 8e-v2 STRC pose-anion-O re-rank stay (P0 — closes the Phase 8g-v2 FAIL margin); Phase 5d ensemble re-dock on Tier-1 holds; Boltz-2 batches closed; v5.3 design freeze still conditional on Tier-1 re-dock + ADMET pass.
Tools & inventory: no new scripts, no new tool cards. Five notes added to vault.

2026-04-26 (02:0x — DR4 v2 prose draft + Stream A/B duplicate-catch + scripts inventory cleanup)

DR4 v2 delivered. File artifacts/deep-research/DR4_v2_route_stratified_safety_case.md — paper-§7-ready route-stratified safety case integrating ALL prior phase findings: route-stratified liability matrix (oral / IT-leak / IT-confined columns); reconciliation with DR5 hydroxamate-otoprotective + Phase 8h-lite 6 Tanimoto-no-overlap + Phase 7I lead-pivot; quantitative anchors table (Givinostat AUC, RWM permeability, Boltz-2 Δ, K1141 distance, ADMET flags, APBS off-target φ); load-bearing §7 caveat = APBS off-target divalent-cation pocket attraction at TRPM4 (+9.36 vs +2.51 kT/e) and TMEM16A (+4.89), CONHOMe-pivot reduces TRPM4 by 3.24 kT/e but doesn’t fully close gate; cochlear melanin / stria vascularis as missed local-route risk; Phase 7J computational sprint scoped (imidazole/triazole ZBG + tetrahydroquinoline core); v3-owed list (PK calc, lit search, bibliography expansion to ~25 sources). Replaces DR4 v1 as the load-bearing safety reference; v1 retained for systemic-route framing.
Streams A/B duplicate-catch. This session also wrote three scripts that duplicate work already executed in this session by parallel Phase 7I work I had not seen: (i) scripts/phase7h_v52_rest11_analyze.py (single-batch v52_rest11) — superseded by phase7i_v52_combined_analyze.py (canonical 12-paired-ligand combined). (ii) scripts/phase7h_v52_top3_analyze.py (single-batch v52_top3) — superseded by same Phase 7I combined. (iii) scripts/phase8h_v53_phosphonate_ensemble.py (2-candidate ensemble check) — superseded by phase7i_tier1_pose_ensemble.py (4-candidate Tier-1 + reference). All three marked superseded in STRC Computational Scripts Inventory § h01 with cross-references; files retained for reproducibility, do not extend.
- One useful cross-check came out of the duplicate runs: the v53A 1-indanyl_phosphonate_-CF3 Vina-pose-anchored min(NZ→ZBG) measurement gave 4.39 ± 0.65 Å in this script vs Phase 7I tier-1’s 4.41 ± 0.86 Å — independent prep, same answer to 0.02 Å. Phosphonate K1141 engagement is reproducible across conformer-seed choice. (CONHOMe-CN diverged: 2.63 vs 4.21 Å — sensitivity to Vina seed on this neutral-head case.)
Stream C state read — Phase 8g-v2 APBS off-target rescore was already complete (5 receptors + STRC ref, all potential.dx files in place since 2026-04-25 night) and had been read out + extended to CONHOMe-Cl during the parallel Phase 7I work earlier this session (phase7i_apbs_offtarget_conhome.py). The remaining queued P1 work is “off-target Vina dock + APBS rescore on the 3 remaining Tier-1 (1-indanyl-phosphonate-CF3, 4-F-biphenyl-phosphonate-CN, adamantyl-CONHOMe-CN)” — DR4 v2 §6 frames the prior assumption that phosphonate selectivity will worsen as a needs-measurement claim, not a settled answer.
Tools & inventory: 3 scripts marked superseded (no-op vs prior table; cross-refs to Phase 7I canonical scripts added). 1 new artifact (DR4_v2_route_stratified_safety_case.md). No new tool cards.
Ranking delta: tier A held | mech 4 held | deliv 4 held | misha_fit 4 held. DR4 v2 is documentation/synthesis, not a proof step. next_step refresh (re-aligned with Phase 7I queue + DR4 v2 §8 / §9): (a) Phase 7J — imidazole/triazole ZBG sprint (P0 light/heavy hybrid: ~6 hr Mac arm64 + Boltz-2 v52_rest11-equivalent on M5 Max) — closes the §6 off-target anti-selectivity caveat with neutral-ZBG bidentate H-bond geometry; (b) Off-target Vina dock + APBS on remaining 3 Tier-1 (P1 light, ~30 min) — completes phosphonate selectivity ranking; (c) TRPM4 Boltz-2 binding-mode test for top Tier-1 — discriminates “real binding” vs “non-binding electrostatic feature”; (d) wet-lab cyto panel + 3-channel patch-clamp + cochlear melanin binding as paper-§7 closure (DR3-vendor-tagged in DR4 v2 §9). Paper draft prose: §3 mech-anchor stays CONHOH; §5 lead pivots to CONHOMe-Cl; §7 selectivity caveat now has APBS quantitative form + concrete chemistry (imidazole/triazole bioisostere, Phase 7J) and wet-lab (3-channel patch-clamp) closure paths. → DR4_v2_route_stratified_safety_case

2026-04-26 (early-AM — Phase 8h-lite #6 Tier-1 refresh; phosphonate prodrug flag raised)

Phase 8h-lite #6 Tier-1 refresh DELIVERED after Phase 7I (parallel agent) closed v52_top3 + pivoted the lead from adamantyl_CONHOH_-Cl (now mech-anchor) to 4 Tier-1 candidates headed by CONHOMe + phosphonate classes. Two ligand-dependent probes (#2 RWM permeability and #5 Tanimoto) re-run on Tier-1; #1 druggability unchanged (pocket-only); #4 pose stability and ADMET re-check already done by Phase 7I work (tier1_pose_stability.csv, tier1_admet_recheck.csv).
Tier-1 RWM verdict: 🟢 adamantyl_CONHOMe_-Cl P 12.9× TMPA → topical-deliverable (best Tier-1, neutral methyl-ester prodrug for CONHOH at pocket); ⚠ adamantyl_CONHOMe_-CN P 3.0× TMPA marginal (TPSA 86.5 + logP 1.44 hurt Avdeef factor); 🔴 BOTH phosphonate Tier-1 BLOCKED at pH 7.4 — 1-indanyl_phosphonate_-CF3 and 4-F-biphenyl_phosphonate_-CN are diprotic with pKa2 ~7.2 → at pH 7.4 ~62% mono-anion + ~38% di-anion + ~0% neutral; HH derate 10⁵-fold; P_eff 10⁻¹³ cm/s (3 orders below TMPA). Prodrug strategy required: bis-POM ester (cf. tenofovir disoproxil), cyclic phosphonate (cf. cyclic-HPMPC), or lipophilic-counterion + nanoemulsion. CONHOH reference unchanged at 7.6× TMPA.
Tier-1 Tanimoto: max 0.154 (aspirin) across 4 Tier-1 + reference, threshold 0.40 → NO Morgan-FP motif overlap with any major ototoxin class on lead-pivot. Aminoglycoside class max ≤ 0.09 across all leads. Lead-pivot CONHOH→CONHOMe/phosphonate introduces zero new ototoxicity-motif risk.
Cross-check vs parallel agent: Phase 7I tier1_pose_stability.csv already showed 4-F-biphenyl-phosphonate-CN at salt-bridge distance 2.83 Å to K1141 NZ (vs CONHOH ref 5.02 Å). The phosphonate’s structural advantage in K1141 engagement is REAL but does not translate to delivery viability without a prodrug. Wet-lab path: bis-POM phosphonate ester + cochlear esterase activation + intratympanic gel formulation (P407 / Mβ-CD per DR5).
Paper-cell update: §6 delivery now reads “CONHOMe-Cl deliverable as topical RWM ear-drop, P 12.9× TMPA. Phosphonate Tier-1 require prodrug formulation.” §3 mech-proof: keep CONHOH-Cl as reference (Phase 5k anchor), pivot CONHOMe-Cl as lead candidate. Speculative: di-anion phosphonate predicted Δ at K1141 NZ ~+9 kT/e (twice CONHO⁻); if APBS confirms, mech 4 → 5 ON PHOSPHONATE COMPOUND, CONHOH stays at 4 for clean disambiguation.
Tools / inventory: 2 new scripts logged (rwm_permeability_tier1.py, tanimoto_ototoxins_tier1.py) in STRC Computational Scripts Inventory § h01. No new tool cards needed (RDKit already documented).
Ranking delta: tier A held | mech 4 held (Phase 5k anchored on CONHOH compound; phosphonate mech-5 candidate awaits APBS) | deliv 4 held but flagged “phosphonate Tier-1 require prodrug” — adds a paper-§6 explicit caveat | misha_fit 4 held. next_step refresh: (1) heavy APBS on CONHOMe + phosphonate at K1141 NZ + on off-target enclosed pockets (parallel agent already running CONHOMe APBS work — phase7i_apbs_work_conhome/); (2) Phase 5d ensemble re-dock on 4 Tier-1; (3) phosphonate prodrug literature review for paper §6 if phosphonate becomes lead. → 6 Tier-1 RWM + Tanimoto refresh 2026-04-26

2026-04-26 (01:30 — DR2 distillation + competitive re-canvass: 3 new atomic notes; landscape extended +3 entries; white-space CONFIRMED)

DR2 enhanced. Re-canvassed the 2024–2026 small-molecule otology landscape against DelveInsight 2026 (“30+ companies / 35+ pipeline drugs”), corporate disclosures, ARO 2026, NIH RePORTER, and DFNB16-targeted searches. Three confirmed missing entries added to §1 master inventory:
- Acousia Therapeutics (ACOU-085 / bimokalner) — first-in-class Kv7.4 K⁺-channel modulator; Phase 2 PROHEAR enrollment completed Jan 2026 (cisplatin ototoxicity); IT slow-release gel; readout Q2–Q3 2026.
- Cilcare CIL001 — cochlear synaptopathy; Phase 2a 2025 (T2D + neurodegen cohorts); €40M Series A Dec 2024; Shionogi partnership May 2025. Cilcare dual-coded as competitor + DR3 CRO.
- Altamira Therapeutics (formerly Auris Medical) — historical/divested reference; AM-111 / AM-101 Phase 3 failures; inner-ear assets sold 2022; pivoted to Bentrio nasal sprays.
Excluded after verification. uniQure/Corlieve (2021 acquisition was AMT-260 temporal-lobe epilepsy, not hearing loss); Novus Therapeutics OP0201 (otitis media via Eustachian tube, Phase 2a missed — not SNHL).
Three DR2-distilled atomic notes created:
- STRC Pharmacochaperone Competitive White Space 2024-2026 — strategic conclusion of canvass: zero small-molecule fold-rescue programs across 17-entry inventory; AAV size barrier (5.3 kb STRC > 4.7 kb cap) protects platform; extends naturally to TECTA/CDH23/MYO7A/OTOA.
- SLC26A4 H723R Pharmacochaperone Precedent — only published otology fold-rescue prior art (DNAJC14 chaperone activation); paper-discussion anchor; H01’s direct-binder mechanism gives defensible novelty vs. machinery-activator approach.
- Otology Funding Pivot to Precision 2024-2026 — capital signal: $1.12 B L i ll y / S e am l ess, €60 MS an o f i / S e n sor i o n,$ 27M Spiral Series B, Otarmeni FDA approval. Pricing benchmarks for STRC Series A; window open through 2026–2027.
Hub evidence list updated with three new wikilinks (post-DR3 block).

Ranking delta

No numerical score change. Tier A · mech 4 · deliv 4 · misha-fit 4 hold.
Strategic positioning reinforced. White-space claim survives a deeper canvass (now 17 entries vs. 14 in DR2 v0). The three new entries (Acousia, Cilcare, Altamira) are otoprotectant / synaptopathy / historical respectively — none touch fold-rescue.
Paper-claim implication. Discussion section now has (a) explicit mechanistic prior art (SLC26A4 H723R / DNAJC14) to cite, and (b) explicit novelty differentiation (direct-binder vs. machinery-activator) for an obvious referee critique.
Pitch-deck implication. Otology funding pivot 2024–2026 ($1.12B Lilly/Seamless ceiling; €40–60M Series A precedents from Sensorion + Cilcare) sets concrete pricing benchmarks for an STRC Series A on Phase 0 fold-rescue confirmation.
Next-step impact. None — heavy queue priorities unchanged (Boltz-2 v52_rest11 analysis already in flight per Phase 7I; APBS off-target pockets queued; v5.3 design conditional on v52_rest11; wet-lab cyto panel as paper deliverable gate).

2026-04-26 (01:5x — Phase 7I follow-on triple delivery: ADMET re-check + pose ensemble stability + APBS off-target on Tier-1)

All three P0 next-steps from the Phase 7I delta closed in one batch.

(1) Tier-1 pose ensemble stability — script scripts/phase7i_tier1_pose_ensemble.py. Generalizes Phase 8h-lite #4 to all 4 Tier-1 + reference. K1141 NZ position from each of 20 Phase 5d MD snapshots; per-ligand min(NZ ↔ ligand-O) across ensemble:

🟢 aq3__4-F-biphenyl__phosphonate__-CN: 2.83 ± 1.33 Å — salt-bridge zone (tighter than reference)
🟢 aq3__adamantyl__CONHOMe__-Cl: 3.89 ± 0.73 Å — salt-bridge zone
🟢 aq3__adamantyl__CONHOMe__-CN: 4.21 ± 0.76 Å — Coulomb sink
🟢 aq3__1-indanyl__phosphonate__-CF3: 4.41 ± 0.86 Å — Coulomb sink
🟢 aq3__adamantyl__CONHOH__-Cl (ref): 5.02 ± 0.57 Å — Coulomb sink (Phase 8h-lite #4 baseline) All 4 Tier-1 candidates engage the K1141 NZ Coulomb sink; two are in TIGHTER direct salt-bridge contact than the original lead. Mech-4 carries through cleanly to the new head-group classes. K1141 NZ rigidity confirmed (radial drift 2.14 ± 0.81 Å). → artifacts/phase7i_v52_combined/tier1_pose_stability.md

(2) Tier-1 ADMET re-check — script scripts/phase7i_tier1_admet_recheck.py. Pure lookup from existing Phase 8f v5.2 ADMET-AI panel; no new computation. 5/5 Tier-1 + reference pass all 10 ADMET gates with 0 flags (clean=True). Head swap from CONHOH to CONHOMe / phosphonate carries through cleanly. Notable percentiles for new Tier-1: hERG 60–76, BBB 12–71, DILI 55–79, CYP3A4 67–88. CONHOMe-Cl best on hERG (75.9) and BBB (70.9). Phosphonates have lower BBB percentile (12–26) — consistent with their −2 charge / membrane-impermeability profile, but BBB is not a relevant gate for intratympanic delivery (per DR5 — apical-turn engagement is the bottleneck, not BLB crossing). → artifacts/phase7i_v52_combined/tier1_admet_recheck.md

(3) APBS off-target rescore — discovery + extension. Phase 8g_v2 (artifacts/phase8g_v2_apbs_offtarget.json) was already run on the CONHOH lead 2026-04-25 night and sat unanalyzed. This session: read out the result + extended the protocol to adamantyl_CONHOMe_-Cl (script scripts/phase7i_apbs_offtarget_conhome.py, output apbs_offtarget_conhome.json). Both runs use APBS 3.4.1 + pdb2pqr 3.6.1 (PARSE force field, NaCl 0.150 M, T=310 K, fine box 28 Å, 65³ grid).

Lead anion-O mean φ (kT/e) at each receptor:

Target	CONHOH	CONHOMe	Δ (CONHOMe − CONHOH)
STRC E1659A K1141 (on-target)	+2.51	+2.48	−0.03
KCNQ4 (7BYM)	+0.91	+1.12	+0.21
TRPM4 (8RD9)	+9.36	+6.12	−3.24
TMEM16A (7ZK3, 6 metals)	+4.90	+4.27	−0.63
Cx50 (8QOJ via Cx36)	−3.02	−3.55	−0.53

The CONHOH lead FAILS the +2 kT/e off-target selectivity gate (TRPM4 +9.36 > STRC +2.51 by +6.85 kT/e; TMEM16A +4.90 > STRC by +2.39 kT/e). This is a load-bearing paper-claim finding that surfaced this session, not before.
CONHOMe head-swap REDUCES off-target electrostatic affinity without losing on-target. TRPM4 drops 3.24 kT/e; TMEM16A drops within strict-gate margin (+1.79 from STRC). Cx50 / KCNQ4 cleanly selective in both runs.
CONHOMe Tier-1 still fails the strict all-targets gate (TRPM4 residual +3.64 kT/e above STRC). Improvement is real but not full rescue.
The 3 remaining Tier-1 (1-indanyl_phosphonate_-CF3, 4-F-biphenyl_phosphonate_-CN, adamantyl_CONHOMe_-CN) need off-target Vina docking before APBS can run on them. Queued as P1 light. Phosphonate (−2 charge) prior is worse off-target electrostatic affinity, esp. at metal-coordinated TMEM16A site — needs measurement, not assumption. → artifacts/phase7i_v52_combined/apbs_offtarget_summary.md
Tools & inventory: 3 new scripts logged in STRC Computational Scripts Inventory § h01 (phase7i_tier1_pose_ensemble.py, phase7i_tier1_admet_recheck.py, phase7i_apbs_offtarget_conhome.py). No new tool cards (APBS / pdb2pqr / RDKit / ADMET-AI all already documented).
Ranking delta: tier A held | mech 4 → 4 (held but rebalanced) — Phase 5k load-bearing on pocket physics is unchanged. CONHOMe Tier-1 has tighter K1141 NZ salt-bridge contact (3.89 Å) than CONHOH reference (5.02 Å) — direct evidence the Coulomb sink mechanism transfers and may be reinforced. The mech 4− qualifier from Phase 7H now resolves to “head-group specific, anti-selective on TRPM4 absent further design.” | deliv 4 → 4 held | misha_fit 4 held | next_step refresh: (a) v5.3 lead pivot now data-grounded across mech (pose-stability) + ADMET + Boltz-2 + APBS axes — aq3__adamantyl__CONHOMe__-Cl is the strongest candidate so far, but TRPM4 anti-selectivity is an outstanding §7 caveat; (b) off-target Vina dock on remaining 3 Tier-1 (P1 light, ~30 min) → APBS rescore on each → completes selectivity ranking; (c) TRPM4 Boltz-2 binding-mode test for top Tier-1 — does the lead actually bind TRPM4 or is the anion-O attractor a non-binding electrostatic feature? (d) wet-lab cyto panel still gates paper. Paper draft prose: §3 mech-anchor stays CONHOH; §5 lead pivots to CONHOMe-Cl; §7 selectivity caveat now has APBS quantitative form + mitigation paths (Boltz-2 binding test, imidazole/triazole bioisostere from DR4). → artifacts/phase7i_v52_combined/{tier1_pose_stability.md,tier1_admet_recheck.md,apbs_offtarget_summary.md}

2026-04-26 (01:3x — Phase 7I v52 combined analysis: v52_top3 closed; lead `adamantyl_CONHOH_-Cl` is a Boltz-2 tie)

Phase 7I closes v52_top3 batch (3 jobs DELIVERED 2026-04-25 night, never analyzed) and unifies it with v52_rest11 into a single 12-paired-ligand picture. Script scripts/phase7i_v52_combined_analyze.py parses both batches, computes per-ligand Δ = mean(E1659A) − mean(WT) ligand_iptm across 5 diffusion samples, and decomposes by body / head / ortho. → artifacts/phase7i_v52_combined/SUMMARY.md
Aggregate over 12 paired ligands: 4 mut-prefer · 5 WT-prefer · 3 tie. Mean Δ = −0.007 ± 0.055, one-sample t = −0.43 (df=11). The “aq3 carrier → mut-prefer” hypothesis is FALSIFIED at the carrier level on v5.2 chemistry — the carrier alone does not pick a side. The signal lives on the head group.
Head-group decomposition (the new finding):
- 🟢 CONHOMe Δ = +0.066 (n=2, sd 0.001) — internally tight; both -Cl and -CN ortho variants mut-prefer
- 🟢 phosphonate Δ = +0.050 (n=2, sd 0.023)
- ⚪ CONHOH Δ = −0.008 (n=2) — the lead head-group is a tie
- 🔴 CONHSO2Me Δ = −0.043 (n=4) — DR4 design-around recommendation falsified for this scaffold
- 🔴 COOH Δ = −0.062 (n=2)
Lead spotlight — adamantyl_CONHOH_-Cl: WT 0.525 ± 0.068, E1659A 0.508 ± 0.114, Δ = −0.017 → ⚪ tie. The v5.2 paper-claim lead is not in the 🟢 mut-prefer cluster on Boltz-2 full-likelihood scoring. Phase 5k decomposed-electrostatics mech-4 proof (formal-anion +4.53 kT/e pocket attraction, p=6.9e-12) holds as a thermodynamic-component result; Boltz-2 picks up the same molecules differently because it integrates entropy + dynamics + non-electrostatic scoring.
New top-tier candidates emerge on adamantyl body: adamantyl_CONHOMe_-Cl Δ = +0.066 and adamantyl_CONHOMe_-CN Δ = +0.065 — methyl-ester hydroxamate prodrug class, both ortho variants mut-prefer. Pairs with rest11’s 1-indanyl_phosphonate_-CF3 Δ = +0.067 (already top in earlier log entry).
Tier reshuffle for v5.3 design freeze:
- Tier-1 (🟢 mut-prefer): aq3__1-indanyl__phosphonate__-CF3 (+0.067), aq3__adamantyl__CONHOMe__-Cl (+0.066), aq3__adamantyl__CONHOMe__-CN (+0.065), aq3__4-F-biphenyl__phosphonate__-CN (+0.035)
- Mech-anchor reference (⚪ tie, role pivot): aq3__adamantyl__CONHOH__-Cl — keep as Phase 5k load-bearing reference compound; demote from “the lead” to “the mech-proof compound”
- Demoted (🔴): COOH and CONHSO2Me heads on aq3-adamantyl body (5 ligands tested, 4 WT-prefer, 1 tie)
Paper framing implication: §3 (mechanism) and §5 (lead) need clean split. CONHOH stays as mech-proof reference (Phase 5k anion-decomposition is its load); CONHOMe / phosphonate becomes the lead-class for binding optimization. ADMET re-check on CONHOMe class is a P0 light follow-up before Phase 5d re-dock — confirms flag-free status carries through the head-group swap.
Tools & inventory: 1 new script logged in STRC Computational Scripts Inventory § h01 (phase7i_v52_combined_analyze.py). No new tool cards (Boltz-2 already documented).
Ranking delta: tier A held | mech 4 → 4 held (Phase 5k load-bearing on decomposed electrostatics; Phase 7I reframes the lead but does not weaken the pocket-physics proof. Mech 4− qualifier from Phase 7H is reinforced — chem-class specificity now resolved one level finer to head-group specificity) | deliv 4 held | misha_fit 4 held. next_step refresh: (1) Phase 5d ensemble re-dock on Tier-1 candidates (4 ligands × WT/E1659A = 8 jobs, ~2 hr compute) — confirms K1141 NZ engagement under MD dynamics, particularly for phosphonate (−2 charge geometry) and CONHOMe (neutral head — different binding-mode class); (2) ADMET re-check on CONHOMe Tier-1 candidates (Phase 8c protocol); (3) APBS off-target enclosed pockets remains P0 heavy unchanged. v5.3 design freeze conditional on Phase 5d re-dock + ADMET pass on CONHOMe class.

2026-04-26 (01:2x — Phase 7H v52_rest11 Boltz-2 analysis DELIVERED; ZBG-pivot evidence for v5.3)

v52_rest11 Boltz-2 batch analyzed. Script scripts/phase7h_v52_rest11_analyze.py parses 20 prediction folders (5 diffusion samples each) from models/boltz_runs/v52_rest11/, extracts ligand_iptm mean ± SD, computes Δ = mean(E1659A) − mean(WT) per ligand variant; positive Δ → mut-prefer = pharmacochaperone signal. 9 complete WT/E1659A pairs out of 11 attempted (2 jobs missing one half: 1-indanyl_CONHSO2Me_-H no-WT, 4-F-biphenyl_COOH_-Me no-E1659A — re-queue if needed).
Aggregate: 3 mut-prefer · 4 WT-prefer · 2 tie. Mut-prefer (ranked):
- 🟢 1-indanyl_phosphonate_-CF3 Δ=+0.067 WT 0.727 · E1659A 0.793 — top hit
- 🟢 adamantyl_CONHOMe_-CN Δ=+0.065 WT 0.534 · E1659A 0.599
- 🟢 4-F-biphenyl_phosphonate_-CN Δ=+0.035 WT 0.537 · E1659A 0.571
Direct DR4 design-around validation:
- Phosphonate ZBG (DR4 Suggestion 2) VALIDATED in 2/2 phosphonate variants tested (1-indanyl-CF3 and 4-F-biphenyl-CN, both mut-prefer). Phosphonate is anionic (−2 at pH 7.4) and predicted by DR4 to provide stronger Coulomb sink than CONHO⁻; Boltz-2 confirms preserved (and stronger) mutant preference. Membrane-impermeability concern is moot since stereocilin is extracellular. Phosphonate eliminates DR4 mutagenicity risk (no Lossen rearrangement) AND maintains K1141 NZ engagement.
- Methyl-ester / hydroxamate prodrug pivot (CONHOMe, neutral) PARTIAL: adamantyl_CONHOMe_-CN Δ=+0.065 mut-prefer; ties to DR4’s prodrug strategy (esterase-cleaved CONHOMe → CONHOH activates locally in cochlea). One geometry, needs more R-substituent variants before generalizing.
- CONHSO2Me (acylsulfonamide ZBG) MOSTLY FAILS with adamantyl tail: 3/4 WT-prefer (Cl, F, H all WT-prefer or strongly so; only CN at tie). Acylsulfonamide does not preserve the K1141 sink in this geometry — exclude from v5.3 design until tested with non-adamantyl tails.
- COOH-F with 1-indanyl WT-prefer Δ=−0.064 — classic carboxylate ZBG insufficient here, consistent with Phase 7H niflumic/mefenamic chemotype-specificity.
Cross-link to DR4 v2 enhancements (artifact DR4_review_and_enhancements.md from earlier this session): the “Computational vetting of design-around proposals” Part F that DR4 review prescribed is now partially executed for ZBG axis — phosphonate validated, CONHSO2Me deprioritized, CONHOMe shortlisted as prodrug pathway. Imidazole/triazole ZBG still not tested; queue for v5.4 batch.
v5.3 design direction (data-grounded, not hand-waved):
- Lead candidate: 1-indanyl_phosphonate_-CF3 (Δ=+0.067 mut-prefer; phosphonate addresses DR4 mutagenicity; 1-indanyl partially saturates aromatic system → addresses DR4 photosensitivity + melanin concerns; CF3 R provides metabolic shielding + lipophilicity tuning).
- Backup: adamantyl_CONHOMe_-CN (prodrug strategy — plasma esterases would expose hydroxamate post-cochlear-deposition; needs cochlear vs plasma esterase differential profiling literature).
- Re-test in Phase 5d ensemble + APBS before any wet-lab move — must confirm phosphonate sustains the K1141 NZ Coulomb engagement under MD dynamics (pose ensemble stability check matching Phase 8h-lite #4 protocol on the new geometry).
Tools & inventory: 1 new script logged in STRC Computational Scripts Inventory § h01 (phase7h_v52_rest11_analyze.py). No new tool cards (Boltz-2 already documented).
Ranking delta: tier A held | mech 4 held (chem-class signal refines but doesn’t elevate to mech 5; phosphonate hit is robustness evidence for the K1141-Coulomb-sink mechanism) | deliv 4 held | misha_fit 4 held. next_step refresh: APBS off-target enclosed pockets (P0 heavy) remains primary, but v5.3 design now has data-grounded direction → Phase 5d-equivalent re-dock + ensemble APBS on 1-indanyl_phosphonate_-CF3 and adamantyl_CONHOMe_-CN is now P0 light (~2 hr compute, validates the v52_rest11 Boltz hits under MD dynamics before chemistry commits). Boltz-2 v52_rest11 analysis closes; Boltz-2 v52_top3 still awaiting comparable analysis (separate prior batch). v5.3 design freeze conditional on phosphonate Phase 5d ensemble check passing. → DR4_review_and_enhancements

2026-04-26 (01:2x — DR3 distillation + improvement plan)

DR3 read in full (artifacts/deep-research/DR3_cro_wet_lab_handoff.md, 149 lines, CRO/wet-lab vendor menu). Content is mostly procurement (vendor lists, FFS pricing, IND-enabling sequence). Per feedback_strc_no_wetlab no vendor/cost content imported into the vault as new notes — instead extracted three durable scientific principles relevant to H01.
3 atomic notes created in ~/STRC/notes/:
- STRC K1141 Fragment Construct Strategy — ~150-aa pocket fragment is the right unit for biophysics AND for the next MD round. Full-length 651k-atom system caps at 2 ns local-overnight (Phase 5d); a fragment system is ~150-180k atoms → ~50-60 ns/day on the same Mac silicon → 100 ns trajectories without GPU rental. AviTag biotinylation pattern + secretion signal recorded for future biophysics.
- Pharmacochaperone Residence Time Criterion — slow k_off (long τ) beats tight K_d for chaperone rescue because the protein needs ER-transit time (tens of min to hours) before ERAD fires. H01 has zero k_off numbers across all phases; every affinity metric to date (Vina, MM-PBSA, APBS, Phase 8h-lite pose stability) is equilibrium-flavored. τRAMD on the K1141 fragment is the cheapest computational route to a relative residence-time ranking.
- Hydroxamic Acid Divalent Cation Liability — CONHOH bidentate Zn²⁺/Ca²⁺ chelation directly hits cochlear ion channels (KCNQ4/TRPM4/TMEM16A/BK/Cx50) and stria vascularis carbonic anhydrases. Phase 8g Vina failure was force-field-blind to metal coordination; APBS-on-Zn-loaded-pockets is the correct rescue. Bioisosteric escapes (CONHSO₂Me, tetrazole) already exist in v5.2 library.
Hub ## evidence updated with three new bullets pointing at the notes.
No score change. mech 4 / deliv 4 / misha_fit 4 hold. The notes surface gaps for the next round of compute, not new evidence.

Ranking delta

tier: A (held)
mech: 4 (held — DR3 distillation reframes the gap as residence-time-not-quantified, but the binding mechanism itself is unchanged)
deliv: 4 (held)
misha_fit: 4 (held)
next_step: heavy queue reordered. Insert two compute jobs surfaced by DR3:
- (2a) APBS-on-enclosed-pockets for off-target selectivity — already queued, but reframe scope around metal-coordination electrostatics (Zn²⁺-loaded pockets), not generic charge complementarity. Direct closure of STRC h01 Phase 8g v5.2 Off-Target Panel 2026-04-25 failure mode.
- (2b) Phase 5l — K1141 fragment AF3 + MD ladder. Score 120/150/180/220-aa fragment boundaries by pLDDT (pocket region) + 50-ns MD K1141 RMSF vs full-length reference. Target ≤ 0.8 Å RMSF in pocket. Output = the construct used for (2c) and any future biophysics.
- (2c) Phase 5m — τRAMD on v5.2 14-compound ADMET-clean shortlist + diflunisal control on the Phase 5l fragment. ~10-20 replicas × 5-10 ns each per ligand. Output = relative residence-time ranking. Reranks shortlist on the actual mechanistic axis.
Existing queue items hold positions: Boltz-2 v52_rest11 batch analysis stays (1); v5.3 design (3) defers until Boltz-2 confirms scaffold rule; wet-lab cyto panel (4) is paper deliverable gate.

2026-04-26 (01:1x — DR4 review + enhancement plan)

DR4 review artifact created. File artifacts/deep-research/DR4_review_and_enhancements.md (~150 lines) — critical review of the DR4 hydroxamic-acid + 2-amino-quinoline-3 IP/tox deep-research file with concrete v2 enhancement plan. Headline: DR4 as written is mis-framed for H01 — it concludes “functionally non-viable without profound modification” from systemic-route assumptions but H01’s planned route is intratympanic. The verdict needs route-stratified rebuilding before any DR4 claim lands in paper §7.
Two direct contradictions with H01 record DR4 must reconcile:
- DR5 hydroxamate-otoprotective SAHA precedent (HSP32 induction, NF-κB / Foxo3a deacetylation block, JNK suppression under aminoglycoside / noise stress) vs DR4 hydroxamate-as-pure-liability framing.
- Phase 8h-lite #5 Tanimoto (Morgan FP r=2 2048b) vs 12-compound ototoxin panel: max 0.127, threshold 0.40, NO chem-class motif overlap — falsifies DR4’s “quinoline-class → retinopathy” structural-analogy inheritance for this lead.
New local-route risk DR4 missed entirely — cochlear melanin / stria vascularis intermediate cells (neural-crest-derived melanocytes, essential for endocochlear potential). Aminoquinoline melanin affinity (chloroquine RPE precedent in DR4 Part C) means intratympanic 2-aminoquinoline could accumulate in the very tissue we need preserved. This is a local-route-specific risk that increases with intratympanic confinement (opposite of systemic risks). Required tests scoped: melanin-Sepharose binding (k_b benchmark vs chloroquine), gerbil cochlea autoradiography, gerbil EP measurements pre/post IT v5.2.
K1141 NZ Coulomb-sink lock-in invalidates drop-in ZBG swap. Phase 8h-lite #4 lead CONHO⁻ ↔ K1141 NZ 5.02 ± 0.57 Å is the load-bearing Mech-4 contact. DR4’s suggested ZBG replacements (imidazole / triazole / carbamoyl-phosphonate) change charge state and either lose the sink (neutral ZBGs) or alter membrane permeability (−2 phosphonate). Each requires Vina re-dock + Phase 5d ensemble re-analysis before being treated as equivalent. New Part F “computational vetting of design-around proposals” prescribed — 1-week sprint, sequenced behind APBS / Boltz-2 v52_rest11.
Quantitative anchors prescribed: Givinostat pediatric oral AUC₀₋₂₄ ≈ 600 ng·h/mL → IT v5.2 systemic AUC projection (expected 10⁻³–10⁻⁴ × Givinostat); melanin binding QSAR; 2-aryl-hydroxamate Lossen stability lit search; CYP3A4 age-stratification for pediatric <2 yr.
Quality issues flagged: bibliography thin (7 unique URLs serving all [1][2][3] citations); blob:capacitor image refs leaked at lines 109, 158, 170 (decode: I_Kr, [hydroxylamine], IC₅₀); IP table row formatting bug at line 49–50 (Pharmacochaperone Mechanism row has only 2 of 4 cells populated); §B.1 has missing adult-data subsection between header and pediatric subsection. Same clipboard-paste artefact pattern as DR5 — single sweep when v2 written.
Ranking delta: no change. tier A held | mech 4 held | deliv 4 held | misha_fit 4 held. Review is survey-quality, not a proof step. next_step unchanged: APBS off-target enclosed pockets (P0 heavy), Boltz-2 v52_rest11 analysis, v5.3 design, wet-lab cyto panel. DR4 v2 sequenced as P1 — behind APBS + Boltz-2 v52_rest11 analysis, ahead of v5.3 design freeze. Currently DR4 cannot be cited in paper §7 without contradicting DR5 + Phase 8h-lite #5; v2 makes it the safety-case spine. → DR4_review_and_enhancements

2026-04-26 (01:1x — DR5 intratympanic delivery deep research ingested + cleaned)

Deep-research artifact DR5 finalized. File artifacts/deep-research/DR5_intratympanic_delivery_state_of_art.md (213 lines) — translational state-of-the-art 2024–2026 for intratympanic / round-window small-molecule delivery, scoped to v5.2 lead (MW 430, log P 1.94, pKa ~9.0, pediatric DFNB16). Original paste contained 17 broken ![Attachment_*.png](blob:capacitor://...) placeholders from clipboard import; decoded and replaced inline: γ-secretase, PPAR-γ, β/Mβ-cyclodextrin, log P, ~37 °C, R-CONHOH, NF-κB, ~70 µm RWM thickness, BLQ. Frontmatter + key-takeaways header + Connections block prepended for vault discoverability.
Key takeaways for H01 formulation:
- Primary translational barrier is BLB vascular clearance, not RWM permeability — apical-turn engagement is the bottleneck (consistent with Phase 8h-lite #2 — basal-turn permeability is the easy half).
- P407 thermogel is the established sustained-release matrix (Otonomy precedent, OTO-104 pediatric safety in NCT02997189) but the lead’s hydroxamic acid (R-CONHOH) chelates Fe(III)/Zn(II) and will catalyze P407 auto-oxidation unless ultra-pure / demetallized / low-peroxide P407 is specified.
- Cyclodextrin guardrail: HP-β-CD / Captisol safe; methyl-β-cyclodextrin (Mβ-CD) catastrophically ototoxic — formulation must explicitly exclude.
- Hydroxamic-acid class is otoprotective (SAHA precedent: HSP32 induction, NF-κB / Foxo3a deacetylation block, JNK suppression under aminoglycoside / noise stress). CONHOH is a feature.
- Pediatric trajectory favorable: RWM more visible in children (Type B/C), thickness ~70 µm age-invariant, tympanostomy-tube delivery enables unsedated quarterly instillation gelling in situ at ~37 °C. Aligns with Misha Compound-Het Therapy Stack Model dosing assumptions.
- IT clinical precedent: broad on safety (OTIVIDEX, OTO-313, PIPE-505, LY3056480, STR001, SENS-401), thin on efficacy — consistent with paper framing that mechanism + delivery viability matter more than first-in-class efficacy hurdles.
Caveat — citation markers [1][2][3][4] unresolved. DR5 ships with claim-level numeric backing (NCT IDs, MW, % response rates, RWM µm) but the bibliography wasn’t attached to the paste. Treat as P1 follow-up before any DR5 fact lands in paper prose; cross-check against ~/BookLibrary/ and PubMed before citing.
Ranking delta: no change. DR5 is translational survey, not a mechanism/deliv proof step. tier A held | mech 4 held (Phase 5k load-bearing) | deliv 4 held (Phase 8h-lite #2 RWM permeability already drove 3→4) | misha_fit 4 held. Strengthens deliv claim with formulation-class precedent + pediatric procedural feasibility but the load-bearing quantitative proof remains Phase 8h-lite #2. next_step unchanged: APBS on off-target enclosed pockets (P0 heavy), Boltz-2 v52_rest11 analysis, v5.3 design, wet-lab cyto panel.
Forward formulation hooks for paper §6 (Delivery): explicit P407 grade spec (demetallized/low-peroxide), explicit Mβ-CD exclusion clause, tympanostomy-tube quarterly dosing schedule, BLB-clearance limitation framing as outstanding-question rather than barrier. → DR5 file.

2026-04-26 (post-midnight — Phase 8h-lite DELIVERED; deliv 3 → 4)

Phase 8h-lite Light Computational Evidence Package DELIVERED. 5 light proofs in <30 s total compute, closing paper-claim gaps using existing Phase 5d/5k/8e data + RAG-mined literature (Salt 2001 RWM, Halgren 2009 SiteMap, Schacht 2008 Auditory Trauma).
- #1 K1141 druggability ✅ — Halgren SiteMap-style Python grid burial estimator (fpocket replacement after Homebrew qhull bug). V 1145 Å³, phobic 0.61, philic 0.39, V_pocket/V_lig 2.75; aromatic-rich shell W1612/F1646/F1169/W1652; K1141 lines 26% of philic shell. All 4 druggability gates pass. → scripts/phase8h_lite/druggability_score.py
- #2 RWM permeability ✅ — Salt 2001 P_TMPA + Stokes-Einstein MW^(−1/3) + Avdeef logP/TPSA + Henderson-Hasselbalch. Lead P 7.6× TMPA → ~60% applied conc in basal turn at 90 min topical RWM application. Sensitivity to CONHOH pKa ±0.5 negligible (7.2–7.6× ratio). Topical ear-drop deliverable. → scripts/phase8h_lite/rwm_permeability.py
- #3 Off-target selectivity (lite) ⚠ INCONCLUSIVE — Vina-pose proxy and receptor-wide K/R scan both fail to discriminate (Vina parks ligand 7.5 Å from K1141 = same Gasteiger artefact; off-target K/R clusters dominate naive count because channels are K/R-rich at voltage sensors / selectivity filters, NOT in enclosed druggable pockets). Proper selectivity test requires APBS on enclosed pockets per off-target — heavy planned step now P0. → scripts/phase8h_lite/selectivity_{charge_proxy,pocket_scan}.py
- #4 Pose ensemble stability ✅ — K1141 NZ position SD 1.11 Å across 20 Phase 5d snapshots (rigid pocket); lead CONHO⁻ ↔ K1141 NZ 5.02 ± 0.57 Å (Coulomb attraction range, edge of direct salt-bridge zone). Mech 4 holds for actual lead geometry, not just generic anion probe. APBS-correct mode would tighten to ~3 Å salt-bridge → projected ~−3 kcal/mol additional binding on top of Phase 5b −8.19. → scripts/phase8h_lite/pose_ensemble_stability.py
- #5 Tanimoto vs ototoxins ✅ — RDKit Morgan FP r=2 2048 bits vs 12-compound class panel (aminoglycoside / loop diuretic / platinum / salicylate / macrolide / quinoline / glycopeptide). Max 0.127 (aspirin), threshold 0.40. NO chemical-class motif overlap. → scripts/phase8h_lite/tanimoto_ototoxins.py
Tools & inventory: 6 new scripts logged in STRC Computational Scripts Inventory § h01. Two new tool cards: fpocket (attempted, broken on Mac arm64 via Homebrew — qhull bug; replacement = our Python burial estimator), RDKit (Morgan FP + descriptors). Skill ~/.claude/skills/alphafold-server/SKILL.md already updated 2026-04-25 with stuck-job recovery; no new skill changes.
Ranking delta: tier A held | mech 4 held (Phase 5k remains load-bearing; #4 reinforces but doesn’t elevate to mech 5) | deliv 3 → 4 (RWM #2 closes “topical ear-drop” delivery claim quantitatively) | misha_fit 4 held. next_step refresh: APBS on off-target enclosed pockets (5 receptors: TRPM4/Cx50/BK/KCNQ4/TMEM16A) is now P0 heavy. Boltz-2 v52_rest11 (DELIVERED 2026-04-25 night, awaiting analysis), v5.3 design, wet-lab cyto panel queued behind. Paper draft prose for sections 1–2, 4–5 can be filled now. → STRC h01 Phase 8h-lite Light Computational Evidence Package 2026-04-26

2026-04-25 (night — literature ingest: Schneider Ed. 2014 De novo Molecular Design)

Literature: ingested De novo Molecular Design (Schneider Ed., Wiley-VCH 2014) — canonical methods textbook for de novo / FBDD / scoring-function classification / efficiency metrics. Source 2014-schneider-baringhaus-de-novo-molecular-design-book; paper note 2014-schneider-de-novo-molecular-design-book. Adds 3 P1 recipes (Recipe — Receptor-Based Scoring Function Selection §1.6.1, Recipe — Fragment Library Filtering Pipeline §5.3.1+§6.2, Recipe — Fragment Optimization Linking Merging Growing §5.3.4+§6.5) + 2 P2 concepts (Fragment Additivity Assumption and Superadditivity §1.5+§5.3.4, Druggability vs Ligandability Distinction §6.3.10) + 3 P0 verbatim tables (Lipinski Rule of Fives vs Congreve Rule of Threes Reference Table Table 5.1, De Novo Design Software Scoring Strategy Catalog Table 1.3, Ligand Efficiency Metrics Catalog §6.4.1). Backs existing audit fixes for Vina (Eq. 1.10 force-field class), MM-GBSA (§16.2.5.2 continuum-correction class), Congreve rule-of-three filter (§5.2.2 / Table 5.1), and LE-based fragment prioritization (§6.4.1 / Eq. 1.8). Sharpens framing: h01 phase 2 / 2b “druggability” scores are properly ligandability scores per §6.3.10 — docstring updates queued. Ranking delta: no tier/mech/deliv/misha_fit/next_step change. Methods foundation strengthened; phase 3c v4 fragment-grow now has a textbook-backed strategy template (growing > linking ≈ merging). Pairs with 2007-chipot-free-energy-calculations-book (theory layer); Schneider 2014 covers the compound-construction + scoring layer above Chipot’s binding-free-energy methods.

2026-04-25 (night — literature ingest: Chipot & Pohorille 2007)

Literature: ingested Free Energy Calculations: Theory and Applications in Chemistry and Biology (Chipot & Pohorille Eds., Springer SCP 86, 2007) — canonical methods reference for h01 phase5 alchemical scoring. Source 2007-chipot-pohorille-free-energy-calculations-book; paper note 2007-chipot-free-energy-calculations-book. Adds 5 P1 recipes (FEP point-mutation algorithm verbatim §2.8.6, soft-core potential §2.8.5 Eq. 2.49, ABF §4.6.4, BAR estimator §5.7.4 Eq. 5.50, LRA pKa method §12.3.2 Eq. 12.40) + 2 P2 concept notes (phase-space overlap diagnostic, single vs dual topology) + 2 P0 verbatim tables. Cross-cuts h26/h09. New literature topic file free-energy-methods. Phase 5 docstrings should now cite Chipot 2007 §2.8.6 + 2.8.5 for the alchemical paradigm and §5.7.4 / pymbar.bar for the estimator. Ranking delta: no tier/mech/deliv/misha_fit/next_step change. Methods foundation strengthened; future Phase 5 alchemical work has a literature-first basis.

2026-04-25 (night — Phase 7H DELIVERED via Boltz-2 local; mech 4 → 4−)

Phase 7H Boltz-2 batch DELIVERED. 12 jobs (6 NSAID × 2 receptors) on Mini-STRC 701 aa via Boltz-2.2.1 on M5 Max GPU (MPS). Wall-time: 1 setup-job (~10 min CCD/weights/MSA) + batch of 11 (1h11m) = ~1h21m total. All 12 succeeded (0 fail). Pipeline: AFS-JSON → phase7h_afs_to_boltz_yaml.py (CCD→PubChem-canonical SMILES via NSAID_SMILES table) → boltz predict on directory → 5 diffusion samples × 12 jobs.
Ligand_iptm WT vs E1659A (mean over 5 models per job):
- 🔴 diflunisal 1FL WT 0.753±0.021 E1659A 0.718±0.023 Δ=−0.035
- 🔴 flufenamic-acid FLF WT 0.782±0.052 E1659A 0.749±0.027 Δ=−0.032
- ⚪ iododiflunisal FHI WT 0.745±0.057 E1659A 0.743±0.025 Δ=−0.002
- 🟢 mefenamic-acid ID8 WT 0.639±0.076 E1659A 0.676±0.055 Δ=+0.037
- 🟢 niflumic-acid NFL WT 0.711±0.095 E1659A 0.766±0.062 Δ=+0.055
- 🔴 tafamidis 3MI WT 0.693±0.054 E1659A 0.669±0.077 Δ=−0.025
- Aggregate: 2/6 mut-prefer · 3/6 WT-prefer · 1/6 tie. No uniform mutant preference.
Interpretation — chem-class signal, not falsification of Phase 5k. Phase 5k measures formal-anion electrostatic Δ (decomposed thermodynamic component, p=6.9e-12 on +4.53 kT/e). Boltz-2 measures full binding likelihood (entropy + dynamics + scoring). They are different quantities. Boltz refines the Phase 5k claim into chem-class specificity:
- Anthranilate-NH-aryl with small ortho-substituent (mefenamic-Me / niflumic-CF₃-on-pyridine) → mut-prefer (+0.04..+0.06)
- Salicylate-COOH (diflunisal / FLF / iododiflunisal) → WT-prefer or tie
- Benzothiazole-COOH (tafamidis) → WT-prefer
- The “any acidic ligand binds via Coulomb” claim is too broad. Geometry of the carrier matters; only certain carriers convert the +4.53 kT/e pocket attraction into a dynamic binding-mode preference.
Ranking delta: tier A held | mech 4 → 4− (chem-class qualifier, Phase 5k stat rigor still valid for what it measures; blanket-acidic claim refined down) | deliv 3 held | misha_fit 4 held. Phase 7H status: DELIVERED (was BLOCKED-EXTERNAL pending AFS allowlist; pivot to Boltz-2 local resolved it). → STRC h01 Phase 7H Boltz-2 NSAID Validation 2026-04-25 (note to be written).
Forward chemistry direction (revised by Phase 7H): v5.2 champion aq3__adamantyl__CONHOH__-Cl is built on 2-amino-quinoline-3 scaffold = an anthranilate-NH-aryl carrier with hydroxamic-acid head — predicts to fall in the 🟢 mut-prefer group per Phase 7H Boltz pattern. Next-step Boltz batch: top-3 ADMET-clean v5.2 (aq3__adamantyl__CONHOH__-Cl, aq3__adamantyl__CONHOMe__-Cl, aq3__1-indanyl__COOH__-Me) × 2 receptors = 6 jobs (~45 min wall) — quick test of the rule on novel chemistry. Defer full 14-compound v5.2 Boltz validation until top-3 confirms the pattern.

2026-04-25 (evening — Phase 7H BLOCKED, redirect to Boltz-2 local)

Phase 7H AF3 NSAID validation BLOCKED. Direct UI walk-through of AlphaFold Server via Playwright MCP confirmed AFS no longer accepts custom small molecules: ligand entity restricted to a fixed dropdown of 19 preset CCDs (ADP/ATP/AMP/GTP/GDP/FAD/NAD/NAP/NDP/HEM/HEC/PLM/OLA/MYR/CIT/CLA/CHL/BCL/BCB) — energy carriers, hemes, fatty acids, citrate, chlorophylls. JSON upload error matrix: custom CCD 1FL → “Ligand is invalid”; SMILES OC(=O)c1cc(-c2ccc(F)cc2F)ccc1O → “Ligand is undefined”. All 6 NSAID CCDs in the prebuilt batch (1FL/3MI/NFL/FLF/FHI/ID8) are non-allowlist. The 12 prebuilt JSONs at ~/STRC/models/af3_jobs_2026-04-24_phase7h/ are unreachable through AFS. Skill ~/.claude/skills/alphafold-server/SKILL.md updated with verified 2026-04-25 flow + allowlist + JS-snippets that don’t depend on shifting ref IDs.
Redirect: Phase 7H pivots from AFS to Boltz-2 local (M5 Max GPU, accepts SMILES, no quota). Same biological intent — independent Vina-free K1141 binding-mode validation — different engine. Bonus: SMILES enables direct test of v5.2 champion aq3__adamantyl__CONHOH__-Cl (it has no CCD, AFS would never have validated it). Tool card to be created at ~/STRC/tools/boltz-2.md before first use per scripts-inventory rule.
Ranking delta: tier A held | mech 4 held (Phase 5k-WT remains the load-bearing mech proof at p=6.9e-12) | deliv 3 held | misha_fit 4 held. Phase 7H demoted from CRITICAL → BLOCKED-EXTERNAL (AFS policy, not biology). No tier change — Phase 7H was always corroboration on top of an already-established mechanism, never a gating proof. Next compute: stand up Boltz-2 + run 14 SMILES (6 NSAIDs + 8 v5.2 ADMET-clean shortlist) × 2 receptors = 28 jobs locally. → STRC h01 Phase 7H AFS Allowlist Block 2026-04-25 (note to be written).

2026-04-25 (pre-dawn — Phase 8g off-target panel + Vina limitation documented)

Phase 8g v5.2 off-target panel: 42 Vina docks (6 ligands × 7 targets: hERG/TMEM16A/KCNQ4/Cx50/TRPM4/COX-1/COX-2). 0/6 pass 100× selectivity gate; even v3b Coulomb-ref fails at TMEM16A (40× stronger off-target). Not a real safety failure — it’s the same Vina electrostatic blindness Phase 5j/4i/5k documented, now symmetrically observed off-target. Vina’s vdW-dominated scoring ranks large hydrophobic membrane-protein pockets as better binders than the shallow polar K1141 pocket regardless of real electrostatic discrimination. Interesting positive: 1-indanyl__COOH__-F rejects COX-2 entirely (ΔG = +1.25 kcal/mol = no binding). Operative off-target safety signals remain: ADMET-AI (Phase 8f) + Phase 6c class liability. Wet-lab patch-clamp + enzyme assay is the decisive test. Ranking unchanged. → STRC h01 Phase 8g v5.2 Off-Target Panel 2026-04-25
Paper draft abstract updated with matched-ensemble Phase 5k-WT numbers (+4.53 kT/e Δ, p=6.9e-12) and v5.2 14-compound shortlist.

2026-04-25 (overnight — Phase 5k-WT matched ensemble + Phase 8d/8e v5.2 running)

Phase 5d-WT MD: 2 ns full-length WT STRC, AMBER14SB/TIP3P/OpenCL, 643k atoms, 14043 s wall (~3h54m at 12.98 ns/day). 20 snapshots at 100 ps intervals. Verified K1141=LYS, E1659=GLU. Completed 2026-04-25 ~01:07 local.
Phase 5k-WT ensemble APBS: 20 WT snapshots × nonlinear PBE (matched protocol to Phase 5k mutant). Pocket mean φ = +1.46 ± 1.55 kT/e (range [−0.81, +4.80]; 16/20 positive). Matched-ensemble comparison to MUT (+5.99±1.37): Δ = +4.53 ± 0.46 kT/e = −2.79 ± 0.28 kcal/mol formal-anion preference for mutant. Welch t=9.81, p=6.9×10⁻¹², Cohen’s d=3.10. Phase 5j static claim (+4.37 kcal/mol) corrected downward 1.57× — WT static was outlier-pessimistic. Specificity claim now paper-ready with statistical rigor.
Ranking delta: tier A held | mech 3 → 4 (matched-ensemble specificity at p<10⁻¹¹ + d=3.1 crosses “mechanism established with statistical rigor”) | deliv 3 held | misha_fit 4 held. → STRC h01 Phase 5k-WT Matched Ensemble APBS 2026-04-25
Phase 8d v5.2 library generation: 384 candidates after drug-likeness filter (MW<500, logP<5, HBA≤10, rotB≤10) over 5 acid heads (CONHOH/COOH/CONHSO2Me/phosphonate/CONHOMe) × 11 tails × 7 R3. Tetrazole SMILES failed kekulization. 280 non-planar + 104 planar. → artifacts/phase8d_v5_2_library/phase8d_v5_2_library.{csv,sdf,json}
Phase 8e v5.2 dock + rescore: 384/384 completed 02:30 local, parallel-4, 2h15m wall. Top combined pre-ADMET: biphenyl__CONHSO2Me__-CN at −8.13; v5 v1 champion adamantyl__CONHOH__-Cl drops to #5 (−8.03) pre-ADMET.
Phase 8f v5.2 ADMET: 10-gate triage on top-30. 15/30 clean. v5 v1 champion resurfaces as #1 ADMET-clean — adamantyl__CONHOH__-Cl is the only candidate at combined < −8.0 AND flag-free. 14-compound shortlist (combined < −7.0 + ADMET-clean): 8 adamantyl + 4 1-indanyl + 2 4-F-biphenyl × 4 acid head classes. → STRC h01 Phase 8d 8e 8f v5.2 Library Design 2026-04-25
Paper draft outline ~/STRC/notes/STRC Paper Draft Outline 2026-04-25.md — nucleation document, 6 sections, ready for prose fill from phase proofs.

2026-04-24 (late night — Phase 8c v5 ADMET triage + overnight queue setup)

Phase 8c — v5 ADMET-AI triage: 20 v5 top combined + 11 v3b refs → ADMET-AI Chemprop ensemble (41 endpoints, DrugBank-percentile gate at 90). v5 top-20: 12/20 flag-free (60%) vs v3b 0/10 (0%). Top combined lead adamantyl__CONHOH__-Cl is fully clean on all 10 gates (hERG 80, BBB 65, DILI 73, AMES, Carcinogens, ClinTox, Bioavailability, CYP3A4, Clearance, LD50). Critical insight: all 1-naphthyl variants fail DILI at 93-95 pct — naphthyl class liability, non-planar cages solve both hERG AND DILI in one move. First ADMET-clean pharmacochaperone candidate in STRC line. Chemistry principle for v5.2: 2-amino-quinoline-3 + adamantyl/BCO/norbornyl + CONHOH + Cl/F/CN. → STRC h01 Phase 8c v5 ADMET-AI Triage 2026-04-24
Overnight autonomous work queued: Phase 5d-WT MD (launched 21:13 local, ~4h Metal-OpenCL) + 3 cron wakeups (00:23 Phase 8d v5.2 library expansion, 03:47 Phase 8f off-target, 07:23 morning report). Fallback: Phase 5d-WT MD runs independently of Claude session via nohup.
Ranking delta: tier A held | mech 3 held | deliv 3 held BUT upgrade path OPENED (ADMET-clean candidate exists → systemic route not pre-closed, pending wet-lab ADME) | misha_fit 4 held.

2026-04-24 (late night — Phase 8 v5 library Coulomb-aware design)

Phase 8a — v5 library generation: 60 ligands around 2-amino-quinoline-3 scaffold (top ensemble-Coulomb from Phase 5k Stage B). Three design axes: non-planar tails (adamantyl / BCO / norbornane / spiro / trans-decalin / cyclohex / indanyl variants — hERG mitigation), CONHOH head (Coulomb-favoring), extended R3 palette (CF3/F/Cl/OMe/CN/H). Drug-likeness filter (MW<500, logP<5.5) + design-priority trim. logP range [−0.5, +2.9] vs v3b top-1 logP 5.47. → STRC h01 Phase 8 v5 Library Coulomb-Aware Design 2026-04-24
Phase 8b — v5 dock + ensemble-APBS rescore: 60/60 docked (Meeko → Vina exh=16 → static E1659A). Pose OA-centroid transplanted into each of 20 Phase 5k APBS grids → ensemble-median Δφ. Combined score = Vina_ΔG + α·ΔG_formal (α=1). Top combined: v5__aq3__adamantyl__CONHOH__-Cl (Vina −6.66 + APBS ensemble −1.53 = combined −8.19 kcal/mol, logP 1.94, 95% pos). Top-20 all non-planar, sp3-rich, combined [−8.19, −6.66] kcal/mol. Distinct chemistry class from v3b — Vina-only iteration would have missed them. Runtime ~20 min parallel-4. → same proof note.
Ranking delta: tier A held | mech 3 held | deliv 3 held (but upgrade-path unlocked via hERG logP 5→2 reduction) | misha_fit 4 held. Chemistry forward path for Misha established.

2026-04-24 (night — Phase 5k ensemble APBS, Tier-0 PASS)

Phase 5k Stage A — ensemble APBS on Phase 5d mutant MD: 20 snapshots × APBS nonlinear PBE (PARSE ff, pocket box re-centred on each snap’s K1141 Cα). Mean pocket φ = +5.99 ± 1.37 kT/e; 20/20 snapshots positive; range [+2.77, +8.70]; median +5.73. Phase 5j static (+5.50) sits exactly at ensemble median — not a pathological frame. ~4 kcal/mol mutant-pocket attraction per unit −1 e anion under dynamics. Verdict PASS. Compute ~25 min (parallel-4 pdb2pqr+APBS). → STRC h01 Phase 5k Ensemble APBS on Phase 5d Mutant MD 2026-04-24
Phase 5k Stage B — per-pose APBS across ensemble: 20 v3b poses × 20 snapshots = 400 measurements via pocket-local Kabsch (static→snap RMSD 1.06 Å mean) + APBS centroid interpolation. Median Δφ = +1.08 kT/e; 92% positive; median formal-anion ΔG = −0.67 kcal/mol per lead-snap. Strengthens Phase 4i single-snapshot result (0.43 → 0.67 kcal/mol, 69% → 92% positive). Top-1 by ensemble Coulomb is 2-amino-quinoline-3__1-naphthyl__CONHOH__-CF3 (−2.86 kcal/mol, 100% positive) — Vina ranked it #8. Vina and Coulomb rankings are nearly orthogonal; v5 library scoring needs combined (Vina_ΔG + α·APBS_ensemble_Δφ) objective. → same proof note.
Ranking delta: tier A held | mech 3 held (Phase 5k confirms Phase 5j at ensemble; mech=4 reserved for holo-MD) | deliv 3 held | misha_fit 4 held. Phase 5g (holo-MD) demoted CRITICAL → INCREMENTAL — direction locked across 4 independent axes (Phase 5j pocket-average, Phase 4i per-pose transplant, Phase 5k-A ensemble pocket, Phase 5k-B pose × ensemble). Holo-MD would only add ligand-dynamics and MM-GBSA decomposition.
Phase 5j residue-ID correction: “D1140/D1173 cluster” phrasing was imprecise — actual nearby acidic residues are D1132, D1146, E1159, E1164 per Phase 5d snapshot content. Mechanism substance unchanged; residue IDs corrected in Phase 5k note + cross-referenced back to Phase 5j.

2026-04-24 (night — Phase 4i APBS pose-transplant, Tier-4)

Phase 4i APBS pose-transplant rescore: 20/29 v3b YELLOW ligands prefer mutant pocket at SAME ligand geometry (pocket-local Kabsch align, 0.21 Å RMSD), median Δφ = +0.70 kT/e, formal-anion ΔG = −0.43 kcal/mol; Vina says 0/29 with +0.55 kcal/mol WT-bias → APBS−Vina gap ~1 kcal/mol. PARTIAL_RESCUE of Phase 5j at per-pose resolution. Naive per-pose (no alignment) gave tepid 15/29 due to frame-mismatch; transplant fixes it. Magnitude ~10× smaller than Phase 5j pocket-average because v3b warhead sits at pocket ENTRANCE not INTERIOR — chemistry direction for v5: push acidic warhead deeper via shorter linker or wrapping tail. Ranking: mech 3 held (Phase 5j already established 3; Phase 4i confirms at per-pose). Tier-4 DONE; Tier-0 ensemble APBS next. Compute: ~15s Python on existing artifacts. → STRC h01 Phase 4i APBS Pose-Transplant Rescore 2026-04-24

2026-04-24 (evening — Phase 5j APBS + Phase 5c-mutant)

Phase 5j APBS WT-vs-mut pocket electrostatics DELIVERED (mech 2 → 3, RESCUE). APBS 3.4.1 nonlinear PBE on identical Phase 4c structures (PARSE ff, pH 7.4, 150 mM NaCl, 310 K, ε_p 2.0 / ε_s 78.54). WT pocket mean φ = −1.60 kT/e (52% positive), E1659A pocket mean φ = +5.50 kT/e (73% positive) — Δ = +7.1 kT/e ≈ +4.3 kcal/mol electrostatic preference for anionic ligands on mutant. Vina doesn’t see this (force-field underweights ionic); APBS does. Phase 4c-v3b WT bias is a Vina artefact, not biology. Mechanism reframe: not “K1141 salt bridge” (Phase 4d falsified that) but “E1659A creates electropositive K1141 pocket from loss of Glu1659 negative charge; acidic ligand binds via net Coulomb potential”. Same prediction, different mechanism. Rescues static pharmacochaperone claim at force-field-independent level. Phase 5g holo-MD + per-frame APBS is now highest-priority test. → STRC h01 Phase 5j APBS WT vs Mutant Pocket Electrostatics 2026-04-24
Phase 5c-mutant cryptic-pocket scan DELIVERED (PASS — K1141 stable on disease target). Grid cavity + RMSF on 20 Phase 5d E1659A full-length snapshots (1-1775, 13,584 heavy atoms/snap, per-snapshot K1141 Cα anchor). K1141 pocket STABLE: RMSF ratio 0.89× global, local void volume 645-770 Å³ across all snapshots. Extends Phase 5c WT Ultra-Mini result to the actual disease target. 10 alt cavities, largest 435 Å³ at 50 Å from K1141 — but matched WT full-length MD missing (Phase 5d ran only mutant), so raw comparison vs Phase 5c WT’s 152 Å³ is not apples-to-apples (Ultra-Mini 701 aa vs full-length 1775 aa). Specificity claim deferred; queued Phase 5d-WT (matched full-length WT MD, 2 ns) as prerequisite for any cryptic-pocket-specificity claim. A held | mech 2 held (from this phase alone). → STRC h01 Phase 5c-Mutant Cryptic Pocket Scan 2026-04-24

2026-04-24 (evening — Phase 6c hERG extension)

Phase 6c hERG extension DELIVERED (FAIL class liability). Added 7th target hERG (human KCNH2) via PDB 8ZYO + XB7 astemizole (Miyashita 2024 Structure 32:1926), box centroid on astemizole. 28 unique ligands docked (10 DEFAULT_PROBES + v3b top-10 + v4 top-10 + 3 ADMET-clean Phase 4h Tier-1). 21/28 hit hERG at <10 µM Kd — every v3b/v4 top-10 tight binder at 0.3-2.4 µM (3-14× tighter than K1141; selectivity h01:hERG 0.07-0.41); 3 ADMET-clean Phase 4h candidates 3-9 µM; tafamidis/diflunisal/benzoxazole-COOH-parents borderline 12-17 µM. ZERO compounds simultaneously clear COX + TMEM16A + hERG + ADMET. Class liability on bicyclic-aromatic-COOH pharmacophore — fenamic / benzofuran / benzoxazole + COOH/tetrazole/CONHOH + lipophilic tail is structurally isomorphic to astemizole-class hERG blockers. deliv 4 → 3 — systemic oral route essentially closed; cochlear intratympanic/RWM delivery now required (was optional). v5 library design must include hERG-selectivity filter (reduce lipophilic tail, basic N introduction, non-planar scaffolds). A held | mech 2 held. → STRC h01 Phase 6c hERG Extension 2026-04-24

2026-04-24 (evening — Phase 7H AF3 validation queued)

Phase 7H AF3 protein-ligand validation QUEUED. 12 AlphaFold Server jobs built in af3_jobs_2026-04-24_phase7h/ — Mini-STRC (1075-1775, 701 aa, K1141→Mini67, E1659→Mini585) × 6 NSAID-class reference ligands (1FL diflunisal, 3MI tafamidis, NFL niflumic, FLF flufenamic, FHI iododiflunisal, ID8 mefenamic — all CCD codes RCSB-verified 2026-04-24) × 2 backgrounds (WT, E1659A). Independent Vina-free test: does AF3 place NSAIDs at K1141 pocket from sequence+CCD alone, and does it resolve a WT-vs-mutant binding-mode difference? Novel Phase 4h combinatorial candidates (benzox-3OH-4F-COOH #15 / benzimid-3OMe-COOH #27 / indole-3OMe #17) NOT submittable — AFS accepts only CCD codes, not SMILES; queued for later local-AF3 / Boltz-2 phase. Submission requires Google auth in browser. Verdict deferred 6-12 h. A held | mech 2 held. → STRC h01 Phase 7H AF3 Protein-Ligand Validation 2026-04-24

2026-04-24 (evening — Phase 7 ADMET-AI triage)

Phase 7 ADMET-AI triage DELIVERED (mixed). ADMET-AI v2.0.1 (ADMET-AI) on 67 unique SMILES (v3b top-30 + v4 top-30 + 21 Phase 4h seeds), 15 medchem gates at 90th-percentile-vs-approved-drugs threshold. 6/67 clean (0 flags), all Phase 4h seeds; 28 light (1-2 flags); 33 heavy (≥ 3 flags). Of 6 clean: 3 fenamic parents already killed by tox audit, and 3 novel candidates — benzox-3OH-4F-COOH (#15), benzimid-3OMe-COOH (#27), indole-3OMe (#17) — are the only compounds simultaneously clearing Phase 6c off-target + scaffold tox audit + ADMET-AI. Tight binders (v3b/v4 Kd ~ 4-5 µM) uniformly flag DILI + solubility + BRENK — affinity-developability trade-off is real. hERG 0/67 and BBB 0/67 above 90th-pct: cardiac + CNS clean on this chemistry class. ADMET-AI flags tafamidis itself at 3 gates (calibration caveat, not absolute kill). A held | mech 2 held. → STRC h01 Phase 7 ADMET-AI Triage 2026-04-24

2026-04-24 (evening — v4 delivery + Phase 4c-v3b + Phase 4d)

Phase 4d K1141A double-mutant decoy DELIVERED (FAIL). LYS1141 → ALA surgical sidechain truncation on E1659A mutant receptor; re-dock 29 v3b YELLOW + 7 Phase 4h Tier-1 / control seeds. Mean salt-bridge loss +0.18 kcal/mol on v3b YELLOW (median +0.26, max +0.65), Tier-1 seeds mean +0.14. 0 of 34 ligands hit the ≥ 2 kcal/mol gate; 4/36 exceed ≥ 0.5; 7/36 actually dock BETTER on K1141A (max gain −0.79 on 2-amino-quinoline-3 + 1-naphthyl + CONHOH + −CF3). The K1141 ε-NH3+ is not the load-bearing anchor the h01 pharmacophore claimed, at Vina force-field level. Scaffold constraint that every lead must carry an acidic warhead is unsupported. CONHOH / tetrazole bioisosteres show steric-interference pattern (Lys sidechain in the way), not salt-bridge pattern. mech 2 held (Phase 4c-v3b already dropped 3→2; Phase 4d independent confirmation). → STRC h01 Phase 4d K1141A Double-Mutant Decoy 2026-04-24
Phase 4c-v3b WT decoy on new chemistry DELIVERED (FAIL). 29 v3b YELLOW + 5 Phase 4h Tier-1 commercial seeds vs WT + E1659A AF3 structures, identical legacy Phase 4c protocol. 0 of 34 acidic ligands prefer mutant; v3b YELLOW mean Δ(WT−mut) = −0.538 kcal/mol (median −0.552, strongest WT-pref −0.989), Tier-1 seeds mean −0.090 (closest to neutral: iododiflunisal-3OMe −0.020, tafamidis −0.061). Legacy Phase 4c had mean −0.455 on 5 leads; v3b chemistry strengthens the WT bias. Higher absolute affinity → stronger WT bias: tight binders (v3b Kd 4–5 µM) are −0.5 kcal/mol WT-biased, weak binders (tafamidis-class Kd ~40 µM) are near-neutral. This falsifies the static geometric pharmacochaperone claim and — crucially — reframes the legacy Phase 4c FAIL: it was never a chemistry-shortlist artefact, it is a property of the AF3 static pockets. Dynamic claim (loop 1642-1651) remains untested. mech 3 → 2 | tier A held. See STRC AF3 Static Pocket Blindness to Loop Dynamics for the methodological framing. → STRC h01 Phase 4c-v3b WT Decoy on v3b YELLOW 2026-04-24
Phase 3c v4 fragment-grow DELIVERED (RED absolute, YELLOW enrichment). 1,572-ligand fragment-grow on v3b winner cluster + Phase 4h playbook seeds. Top mean ΔG −7.27 kcal/mol (Kd 4.65 µM, f_PC 0.341) — within 0.01 kcal/mol of v3b ceiling (−7.28, 4.57 µM, 0.343). 0 GREEN / 32 YELLOW / 18 RED. Fragment-grow did NOT break the K1141 pocket affinity ceiling; local chemistry on the 3-amino-benzofuran-2-COOH + tetrazole/CONHOH + naphthyl-biphenyl class is saturated. Phase 4h tafamidis-class seeds contribute only 1/top-20 — bioisosteric seeds do not out-dock the combinatorial on Vina. Further chemistry gain requires either (a) reversible-covalent Lys warhead (Phase 6b), (b) cryptic-pocket discovery on mutant trajectory (Phase 5c-mutant), or (c) ligand-bound MD-resolved pocket for rational growth. A held; next_step updated to Phase 5g holo-MD after Phase 4c-v3b/4d mechanism tests. active_runs cleared. → STRC h01 Phase 3c v4 Fragment-Grow Delivery 2026-04-24

2026-04-24 (afternoon — Phase 6c full 6-target panel)

Phase 6c EXTENDED to 6/6 dockable targets (4 min 42 s wall, 60 docks). Bound-ligand PDBs verified via WebFetch (no fabricated pocket coords): TRPM4 8RD9/A1H0C anthranilic blocker (Ekundayo 2025), Cx50-via-Cx36 8QOJ/YMZ mefloquine (Ding 2024, cross-paralog flag), KCNQ4 7BYM/FBX retigabine (Li T 2021), TMEM16A 7ZK3/JRF 1PBC pore blocker (Lam 2022), COX-1 1Q4G/BFL, COX-2 5IKR/ID8. Old guesses replaced (TRPM4 7MF0=404, Cx50 7JN0=apo, KCNQ4 7BYL=KCNQ1, TMEM16A 5OYB=apo). BK skipped (no NSAID-bound PDB exists in RCSB; relegated to Phase 8 G3 patch-clamp). Script extended with cif-fallback download + cif HETATM centroid parser (obabel cif→pdb conversion drops HETATMs). REAL FINDINGS: (1) TMEM16A is a SECOND off-target wall — tafamidis Kd 4.7 µM, benzoxazole-3CN 2.9 µM, benzoxazole-3OMe 2.6 µM; bicyclic-COOH pharmacophore engages 1PBC pore-blocker site; expressed in stria vascularis = ototoxicity risk. (2) KCNQ4 cleanly de-selected for ALL benzoxazole bioisosteres (Kd 442-559 µM) — the most safety-critical target (DFNA2 paralog) is clean. (3) TRPM4 mildly de-selected (3-4× vs flufenamic) — borderline at 10 µM dose. (4) Cx50 (Cx36 proxy) inconclusive due to cross-paralog uncertainty. EVERY Phase 4h Tier-1/Tier-2 candidate has ≥3 sub-10 µM off-targets. Material constraint: no current playbook compound passes 3-of-3 selectivity gate (h01 + COX + TMEM16A). Action: cross-dock v4 top-30 against this panel post-verdict (~5 min wall) BEFORE Phase 8 ordering; queue v4b carboxylate-deselect sub-strategy if v4 inherits same off-target pattern. A held; Phase 8 SOP needs TMEM16A as G2.5 kill alongside COX. Pipeline calibration: COX within 2× of published IC₅₀s, channels show 5-19× site-specific deviation (use rank-order not absolute Kd for channels). → STRC h01 Phase 6c Full 6-Target Panel 2026-04-24

2026-04-24 (afternoon — Phase 6c COX selectivity)

Phase 6c COX-1/-2 selectivity panel DELIVERED (1m 36s wall, 10 ligs × 2 dockable targets out of 7 planned). Stage A receptor prep: COX-1 (1Q4G) + COX-2 (5IKR) dockable via bound-NSAID centroids (BFL / ID8); TRPM4 7MF0 fetch failed (wrong PDB ID); Cx50/BK/KCNQ4/TMEM16A apo cryoEMs require lit-curated boxes (deferred per literature-first discipline). Stage B: pipeline calibrated — niflumic Kd 0.28 µM COX-2 / 0.42 µM COX-1, flufenamic 0.37 / 0.51 µM, both within 1 order of magnitude of published IC₅₀s. REAL FINDING: tafamidis Kd 1.5 µM COX-2 / 1.5 µM COX-1; benzoxazole-3CN-COOH (Phase 4h Tier-2 lead) Kd 1.8 / 1.0 µM; benzoxazole-3OMe-COOH 0.99 / 0.97 µM. Phase 4h “bicyclic core kills COX” design thesis FALSIFIED on docking — Arg120 salt bridge + bicyclic-COOH pharmacophore preserved at COX. Selectivity vs h01 mutant ensemble: parent fenamates prefer COX 88-147× over K1141. Material design constraint: any v4 lead carrying COOH + bicyclic aromatic needs explicit COX-deselect sub-step (e.g., replace COOH or add Tyr385-clashing group). Action items logged in proof note (TRPM4 PDB replacement + 4 lit-curation tasks). A held; Phase 8 SOP COX should be promoted from G3 to G2.5 kill-gate. → STRC h01 Phase 6c COX Selectivity Panel 2026-04-24

2026-04-24 (afternoon launch — Phase 3c v4)

Phase 3c v4 LAUNCHED (PID 81208 nohup, 16:47 local, ETA ~17:57 local = ~70 min wall on Mac CPU=8). Library: 1,572 ligands fragment-grown on v3b YELLOW winner cluster + Phase 4h tafamidis-playbook seeds. Locked cores: 3-amino-benzofuran-2-COOH (v3b dominant), 2-amino-benzoxazole-5-COOH (Phase 4h core A, tafamidis class), 2-amino-benzothiazole-6-COOH (Phase 4h core B). Distal aryls (24): Phase 4h polar set (CN/OMe/NHSO₂Me/OH-F/F + naphthyl/biphenyl + heteroaryl). Ring subs (6): -H/-F/-CF₃/-OMe/-OH/-Me. Acid bioisosteres (4): COOH/CONHOH/CONHSO₂Me/tetrazole (CONHOH dominated v3b top-10). Plus 22 explicit Phase 4h seed SMILES (incl. tafamidis parent + iododiflunisal + benzoxazole-3CN/3OMe variants + 3 v3b top-3 reference replays). 2-stage ensemble dock identical to v3b (Stage 1 snap_008 exh=8 × all; Stage 2 top-50 × 5 k-means conformers exh=16 × 5 modes). SMOKE PASSED (30 ligs × 1 receptor in 66 s; top hit −7.04 kcal/mol on benzofuran-2-COOH + 3-CN-phenyl + CONHSO₂Me + −CF₃ — non-fenamic, COX-broken, TRPM4-killing combo per Phase 4h tox-breaker design). A held; verdict deferred. → STRC h01 Phase 3c v4 Fragment-Grow (write-up pending on completion)

2026-04-24 (afternoon delivery)

Phase 5e mutant ensemble re-dock DELIVERED (10 h wall, 19:36 → 05:35 local). 11 ligands × 20 Phase 5d E1659A MD snapshots × Vina exh=16, per-snap K1141+ring-Cα centroid box (18 Å). Pocket-stability gate PASSED: 6/11 WT-comparable ligands max kd_ratio 2.09× (diflunisal), mean 1.49× — under +1 kcal/mol ΔΔG threshold. WT-based docking validated as proxy for E1659A → v3b 29 YELLOWs do not need re-docking on mutant; Phase 5f deferred. Chemistry on mutant ensemble: 0 GREEN / 0 YELLOW / 11 RED at MILD-MODERATE f_PC ≥ 0.30. Best mutant lead flufenamic Kd 32.7 µM f_PC 0.117 (fenamic-tox blocked, MD probe only); tafamidis-analog Kd 51.3 µM f_PC 0.082 — bioisosteric pivot intact but single analog insufficient. Best snapshot ΔG −7.20 (flufenamic), −7.11 (tafamidis-analog) confirms upper-tail capacity exists. A held; next_step → Phase 3c v4 fragment-grow on 3-amino-benzofuran-2-COOH scaffold (Phase 4h Tier-1 commercial seed set: tafamidis, iododiflunisal, benzoxazole-3CN, benzoxazole-3OMe). active_runs cleared. → STRC h01 Phase 5e Mutant Re-Dock Delivery 2026-04-24

2026-04-25

Halgren 2009 SiteMap full paper retrieved via Anna’s + parsed; SiteScore/Dscore formulas extracted; druggability docstrings verified against primary source.
Citation audit: 6 ⚠ rows in literature/pharmacochaperone.md closed to ✅; 3 paper notes created (Eberhardt 2021, Genheden 2015, Halgren 2009); Vina gate / MM-PBSA gate labeled pipeline-specific with method citations; MW/TPSA context labeled; druggability docstrings updated across 6 scripts; Halgren 2009 JCIM downloaded via Anna’s Archive (MinerU parsing in progress). A held. → STRC h01 Parameter Provenance Audit 2026-04-25

2026-04-24

Phase 3c v3b + 6b: 0 GREEN / 29 YELLOW / 21 RED on 12,253-ligand ensemble dock. Ceiling mean ΔG −7.28 kcal/mol (Kd 4.57 µM, f_PC 0.34) — 6.6× lower Kd than v2 fenamic (30 µM) via 3-amino-benzofuran-2-COOH scaffold. Crosses MILD-MODERATE adjunct threshold (f_PC 0.30), short of NORMAL (0.50). Top YELLOW cluster tight on benzofuran-2-COOH body + naphthyl/biphenyl tails + CONHOH/tetrazole acid bioisostere + −CF₃. A held. → STRC h01 Phase 3c v3b + 5d Delivery 2026-04-24
Phase 5d E1659A full-length MD: delivered 2 ns / 20 snapshots (not planned 10 ns; 14.06 ns/day Metal-OpenCL, production-10ns-ETA was 17.1 h). First MD on actual disease target (1,775-res AF3 E1659A, 651k-atom solvated); closes WT-Ultra-Mini gap. K1141/E1659 mutations verified. A held. → STRC h01 Phase 3c v3b + 5d Delivery 2026-04-24
Phase 5e mutant-ensemble re-dock LAUNCHED: 11 ligands (LEGACY_LEADS + V2_HITS + tafamidis-analog) × 20 Phase 5d mutant snapshots × Vina exh=16, box centre per-snapshot from K1141+ring Cα centroid, 18 Å box. Gates v3b YELLOW-on-mutant validity. ~15 min wall. A held, next_step updated.

2026-04-23

Phase 4h Tafamidis-Playbook Library drafted: 30-compound prioritized target list, 5 cores × 6 polar distal subs × 2 acid bioisosteres = 60 combinatorial pruned to top 30. Tier 1 commercial + Tier 2 2-step in-house = 8 compounds orderable this week (tafamidis #5, iododiflunisal 12, benzoxazole-3CN #1, benzoxazole-3OMe #2). 3 Phase 6b covalent warheads (salicylaldehyde imine, α-cyanoacrylate Michael, acyl-hydrazone). Predicted ΔG improvement -0.5 to -2.0 kcal/mol over parent niflumic −6.18. 6-target cochlear channel off-target panel pre-scored (TRPM4/Cx50/BK/KCNQ4/TMEM16A/COX) queued as Phase 6c 30×6 = 180 docks, ~1 h local wall. A held. No empirical data yet; this seeds Phase 3c v4 virtual screen + Phase 8 wet-lab synthesis Stage 1. → STRC h01 Phase 4h Tafamidis Playbook Library 2026-04-23
Phase 5d E1659A MD LAUNCHED (PID 28875 nohup, 19:34 local 2026-04-23, ETA ~23:00 local = 3.5 h wall): full-length AF3 mutant job3-mutant.cif chain A (1775 residues), verified K1141=LYS + E1659=ALA pre-run. 645,244-atom system (4× Phase 5a) solvated with TIP3P + 0.15 M NaCl, AMBER14SB, 2 ns production × 20 snapshots. SMOKE (50 ps × 2) passed at 13.86 ns/day. Closes critical gap: all prior Phase 5a/5b/5c were on WT Ultra-Mini (594-1294 full-length) which excludes E1659. Phase 5d is the first MD on the actual disease target. Phase 5e script written (mutant-ensemble Vina re-dock of LEGACY_LEADS + V2_HITS + tafamidis-analog, pocket box centre derived per-snapshot from K1141 + ring residue Cα centroid, 18×18×18 Å box, cpu=8 parallel). Phase 5e scheduled post-Phase 5d delivery + post-v3b Stage 2 (CPU schedule). Verdict deferred. → proof note STRC h01 Phase 5d E1659A MD 2026-04-23 (TODO, pending delivery)
Fenamic Scaffold Tox Audit: A held, scaffold-developability risk sharpened. Parent fenamates (niflumic / flufenamic / meclofenamic / mefenamic / tolfenamic) NOT developable as-is for pediatric cochlear target — ion-channel promiscuity (TRPM4 2.8 µM, Cx50 3 µM gap junctions, BK 25 µM, KCNQ-family, TMEM16A 12 µM) overlaps Kd ~30 µM window; mefenamic seizure risk ≥2.5 g; COX inhibition reduces cochlear blood flow (therapeutically antagonistic). Scaffold developable via tafamidis-style optimization (benzoxazole bioisostere + polar distal ring subs); direct precedent flufenamic → tafamidis → FDA 2019. Phase 4h: parents as MD PROBES only, not wet-lab triage candidates. → STRC h01 Fenamic Scaffold Tox Audit 2026-04-23
Phase 8 Wet-Lab Triage SOP drafted: 3-gate protocol ready for GREEN-scenario from v3b. G1 ThermoFluor ΔTm (iododiflunisal positive, 3-day ~ $3 k), G 2 MST K d (2 - w k$ 8k), G3 cryoEM + HEK293-FL-E1659A-GFP surface trafficking rescue (4-wk ~$25k) + parallel cochlear ion-channel tox pre-screen (TRPM4, Cx50, BK, KCNQ4, TMEM16A). Decision tree through Phase 9 mouse OHC ex-vivo + CRO. → STRC h01 Phase 8 Wet-Lab Triage SOP
Phase 3c v3b + 6b LAUNCHED (in progress, PID 73601 nohup detached, ETA ~5.7h): fenamic-focused 12,253-ligand library (619-lig combinatorial → expanded to 12k via 8 cores × 57 N-aryl subs × 5 acid bioisosteres × 8 ring subs + 6 covalent warheads on 8 cores × 10 hot N-aryls × 2 acid bioisosteres). Two-stage ensemble dock identical to Phase 3c v2b: Stage 1 on snap_008 exh 8 × 3 modes (0.65 lig/s, ETA 313 min), Stage 2 on top-50 × 5 k-means-selected conformers exh 16 × 5 modes. Checkpoint JSON every 100 lig. Expected delivery ~04:30 local. Output → pharmacochaperone_phase3c_v3b_ensemble_dock.json. Will produce proof note STRC h01 Phase 3c v3b Fenamic + Covalent Screen 2026-04-23 on completion. → STRC h01 Phase 3c v2 Expanded Screen 2026-04-23 (precedent)
Phase 3c v2 expanded virtual screen: RED overall, partial progress. 667-ligand library (619 RDKit combinatorial scaffold × acid bioisostere + 29 curated FDA/literature drugs) × 2-stage ensemble docking (Stage 1 snap_008 exh 8 on all 667 → 8.5 min; Stage 2 top-30 × 5 k-means conformers exh 16 → 4 min, total 13 min wall). Top hits niflumic-acid / flufenamic-acid / sulfasalazine at Kd ≈ 30 μM, f_PC ≈ 0.125 at [L]=10 μM — 1.7× better Kd than Phase 5b diflunisal baseline. Fenamic-acid family (2-arylaminobenzoic) = new scaffold direction; tetrazole bioisostere competitive with COOH. 0 GREEN / 0 YELLOW / 30 RED. Ceiling analysis: to cross f_PC ≥ 0.30 (MILD-MODERATE) need ΔG ≈ -7.06 kcal/mol (0.88 below current best). A held, all scores unchanged. Next-step reordered: (1) Phase 3c v3 ZINC22 bioactives ~20k library, (2) Phase 3c v4 fragment-growing niflumic core, (3) Phase 6b reversible covalent Lys warhead. → STRC h01 Phase 3c v2 Expanded Screen 2026-04-23
Phase 5c cryptic pocket analysis: K1141 site GREEN-LIGHT. Custom grid-cavity + Kabsch-aligned RMSF analysis on Phase 5a 20-snapshot trajectory (fpocket 4.0 brew build broken on qhull). K1141 pocket Cα RMSF 0.62 Å vs global 1.23 Å (2× more rigid than average); local void volume 719-850 Å³ across 20 frames (~15% CV, pocket does not collapse or shift geometry). Global cavity scan (snap_010, 26-direction burial ≥16/26): no alt cavity > 152 Å³, nearest at 18.7 Å (91 Å³). Phase 5b RED-LIGHT confirmed chemistry-limited not site-limited. Phase 3c v2 expanded screen targeted at K1141 with ensemble receptor docking (use all 20 Phase 5a snapshots) validated as next move. A held, all scores unchanged. → STRC h01 Phase 5c Cryptic Pocket Analysis 2026-04-23
Phase 5a MD pipeline validated on local Mac: 62 ns/day OpenMM OpenCL (Metal backend) on 164 k-atom solvated Ultra-Mini × TMEM145 chain A. 2 ns production delivered 20 snapshots, 45 min wall time. Replaces deferred GROMACS/AmberTools scaffold which required A100 rental. Phase 5b Vina ensemble re-docking of Phase 4b top-5 leads + diflunisal positive control against 20 snapshots → all f_PC at [L]=10 μM < 0.10 (best = diflunisal positive 0.083, best lead = naphthalene-2-COOH 0.047). Phase 4b single-structure was over-optimistic by +0.36 to +0.92 kcal/mol. RED-LIGHT: current shortlist insufficient for NORMAL-rescue monotherapy. A held, scores unchanged (mechanism intact, Misha fit intact; only shortlist insufficient). Next step: Phase 3c v2 expanded virtual screen (DrugBank FDA + ZINC22 carboxylate tranche + fragment-based; ensemble filter ΔG ≤ −7.5 kcal/mol). → STRC h01 Phase 5 MD Ensemble Rescoring 2026-04-23
Misha compound-het stack integration: A held, scores unchanged, framing strengthened. Per Misha Compound-Het Therapy Stack Model binary-functional-OHC model, h01 PC is the only monotherapy route to NORMAL (≤25 dB ABR) for Misha — because PC reaches every OHC (including non-transduced ones where h03 AAV by definition does nothing), rescuing maternal E1659A to 0.5 × f_mat_treated ≥ θ. Required f_PC ≥ 0.50 (mild E1659A) to ≥ 0.75 (severe). Phase 5 MD ensemble MM-GBSA priority strengthened — f_PC is the critical-path deliverable. → Misha Compound-Het Therapy Stack Model
Post-audit housekeeping: A held. TPSA docstring in phase3b_virtual_screen.py corrected (descriptor-only, not composite scoring; CNS TPSA bracket removed — STRC extracellular). Druggability cross-phase incomparability flagged in Phase 1/2/2b. No fabricated ranges added. → STRC h01 Parameter Provenance Audit 2026-04-23

2026-04-22

S → A: mechanism 4→3; STRC extracellular protein, 4PBA ER mechanism mismatch → STRC Repurposed FDA Chaperone Branch
Phase 4g: 4PBA first mut-preferring compound (−0.32 kcal/mol); repurposing branch opened then killed same day → STRC Pharmacochaperone Phase 4g Repurpose Screen
Phase 4f SMOKE: ΔΔG 20,439 kcal/mol = method error; production canceled → STRC Pharmacochaperone Phase 4f Interface Rescue SMOKE

2026-04-21

Phase 4a: PASS 4/5, K1141 pocket reproduces across CIFs → STRC Pharmacochaperone Phase 4a Pocket Reproducibility
Phase 4b: PASS on LE; leads beat diflunisal 29-57% → STRC Pharmacochaperone Phase 4b Smoke Test
Phase 4c: FAIL; all 5 leads prefer WT (mean −0.455 kcal/mol); aspect (a) flagged → STRC Pharmacochaperone Phase 4c WT Decoy
Phase 4e: soft-FAIL margin 0.15 vs 0.20; re-opens as Phase 4b sub-test → STRC Pharmacochaperone Phase 4e Off-Target Selectivity
Phase 4 plan written, 7-gate ladder → STRC Pharmacochaperone Phase 4 Plan
Phases 0-3 complete; top-5 leads shortlist → STRC Pharmacochaperone Virtual Screen Ranked Leads

2026-04-23

#1 Pharmacochaperone Phase 4h Tafamidis-Playbook Library drafted: 30-compound bioisosteric optimization seed list — 5 core scaffolds (benzoxazole / benzothiazole / benzimidazole / diflunisal-iodo / indole-3-COOH) × 6 polar distal-ring subs (CN / OMe / OH-F / CF₃ / NHSO₂Me / F) × 2 acid bioisosteres (COOH / acyl-sulfonamide), pruned from 60 combinatorial. Tafamidis parent + iododiflunisal as Tier 1 positive benchmarks. 3 Phase 6b reversible-Lys covalent warheads (salicylaldehyde imine, α-cyanoacrylate Michael, acyl-hydrazone). Ranked by synthesis tractability (Tier 1-5, 8 commercial/in-house, budget est $150-250k for 21-compound 8-wk synthesis ladder). Tox-breaker design: kill COX via bicyclic heteroaromatic core + kill cochlear channel panel (TRPM4/Cx50/BK/KCNQ4/TMEM16A) via distal halide→polar swap. Seeds Phase 3c v4 virtual screen + Phase 8 synth Stage 1 (parallel tracks). A held. → STRC h01 Phase 4h Tafamidis Playbook Library 2026-04-23
#1 Pharmacochaperone Phase 5d LAUNCHED (PID 28875 nohup, started 19:34 local 2026-04-23, ETA ~23:00 local = 3.5 h wall): E1659A mutant MD on full-length AF3 prediction job3-mutant.cif chain A (1775 residues, verified K1141=LYS and E1659=ALA before start). 645k solvated atoms (4× Phase 5a), AMBER14SB + TIP3P identical to Phase 5a for comparability, 2 ns production × 20 snapshots. SMOKE passed at 13.86 ns/day (50 ps + 2 snapshots in 17 min). Closes methodology gap — Phase 5a/b/c all ran on WT Ultra-Mini (594-1294 full-length = residues 1-701 in construct) which does NOT contain E1659. Phase 5d re-opens the question of whether mutation distorts K1141 pocket geometry enough to invalidate WT-based docking. Companion Phase 5e script written (mutant-ensemble re-dock of LEGACY_LEADS + V2_HITS + tafamidis-analog positive, scheduled post-Phase 5d + post-v3b Stage 2). Verdict deferred. A held. → proof note STRC h01 Phase 5d E1659A MD 2026-04-23 (TODO, to be written on delivery)
#1 Pharmacochaperone Fenamic Scaffold Tox Audit: A held, scaffold-developability risk sharpened. Literature audit (PubMed + LiverTox + meta-review PMC12530847) on parent fenamates nominated by Phase 3c v2 (niflumic / flufenamic / meclofenamic / mefenamic / tolfenamic). Parent fenamates are NOT developable for 8-y-o DFNB16 patient as-is: (a) pan-ion-channel promiscuity — TRPM4 IC₅₀ 2.8 µM, Cx50 3 µM (cochlear gap junctions), BK 25 µM, KCNQ-family 56 µM (KCNQ4 = DFNA2 HL paralog), TMEM16A 12 µM — overlaps our Phase 3c v2 Kd ~30 µM; (b) mefenamic CNS seizure signal at 2.5 g overdose threshold (Kamour 2017 PMID 28066875); (c) COX inhibition reduces cochlear blood flow — directly antagonistic to therapeutic goal. But scaffold IS developable via tafamidis-style bioisosteric optimization — direct precedent: flufenamic → tafamidis (Pfizer, FDA Vyndaqel 2019, TTR pharmacochaperone, pediatric-cleared). Design constraints for Phase 4h/7 medchem: (a) swap carboxylate → benzoxazole bioisostere (kill COX pharmacophore); (b) replace distal aryl halides (CF₃, diCl) with polar small groups (CN, OMe, F) to kill TRPM4/Cx50/BK affinity. Action for Phase 4h: use niflumic/flufenamic/sulfasalazine as chemical PROBES for MD-based pocket validation only, NOT wet-lab candidates. Hand pharmacophore to medchem after probe validation. → STRC h01 Fenamic Scaffold Tox Audit 2026-04-23
#1 Pharmacochaperone Phase 8 Wet-Lab Triage SOP drafted: A held. 3-gate wet-lab validation protocol prepared in advance of Phase 3c v3b GREEN-scenario arrival. Gate 1 — ThermoFluor ΔTm (Sf9-expressed mini-STRC-UM-WT + mini-STRC-UM-E1659A, extended to 594-1775 to include mutation site; iododiflunisal positive control, ≥1.5 °C shift = PASS on mutant at 100 µM; 3-day, ~ $3 k) . G a t e 2— MST K d + s t o i c hi o m e t ry (N an o T e m p er M o n o l i t h, RE D - t r i s - NT A o n H i s 6, 16 - ptt i t r a t i o n 500µ M \to 15 n M a t 25° C an d 37° C, P A SS K d \leq 50µ M + H i ll 0.8 - 1.2; 2 - w ee k,$ 8 k). Gate 3 — cryoEM co-complex (K3 or Falcon 4, ligand density ≥4σ at K1141 pocket, modeled pose within 1.5 Å of Vina pose) + HEK293-FL-E1659A-GFP surface trafficking rescue (flow cytometry, PASS ≥30% surface at 10 µM vs 15% baseline; 4-week, ~$25 k). Parallel tox pre-screen patch-clamp TRPM4/Cx50/BK/KCNQ4/TMEM16A kill >25% inhibition at 10 µM. Expected winnowing 20 → 3–6 → 1–3 → full-pass. Phase 9 = mouse OHC ex-vivo + CRO IND-enabling tox. → STRC h01 Phase 8 Wet-Lab Triage SOP
#1 Pharmacochaperone Phase 3c v3b + 6b LAUNCHED (in progress, PID 73601 nohup detached, ETA ~5.7 h from 10:49 UTC launch → completion ~04:30 local 2026-04-24): 12,253-ligand fenamic-focused library (8 anthranilic/fenamic core scaffolds × 57 N-aryl substituents × 5 acid bioisosteres × 8 ring subs + Phase 6b covalent warhead variants: 8 cores × 10 hot N-aryls × 6 warheads — acrylamide / α-cyanoacrylate / β-ketoamide / α-ketoamide / salicylaldehyde-like / α-chloroacetamide — × 2 acids). Two-stage ensemble dock (Stage 1 all-lig × snap_008 exh 8; Stage 2 top-50 × 5 k-means conformers exh 16). Checkpoint JSON every 100 lig. Pre-launch expectations: fenamic-scaffold expansion could push best Kd from Phase 3c v2 niflumic 29.5 μM toward 5-15 μM (GREEN band if σ_ensemble holds); covalent warheads potentially GREEN if reversible Lys adduct viable. Verdict deferred pending completion. Status: no ranking change yet. → STRC h01 Phase 3c v2 Expanded Screen 2026-04-23 (precedent)
#1 Pharmacochaperone Phase 3c v2: RED overall, partial progress, ceiling apparent. 667-ligand expanded screen (619 RDKit combinatorial scaffold × acid bioisostere + 29 curated FDA carboxylates) against K1141 with two-stage ensemble docking (Stage 1: snap_008 global rep exh 8; Stage 2: top 30 × 5 k-means conformers exh 16). Top 3: niflumic-acid 29.5 μM / flufenamic 30.0 μM / sulfasalazine 30.7 μM → f_PC ≈ 0.125 at [L]=10 μM, η=0.5. 1.7× better Kd than Phase 5b baseline (diflunisal 50 μM). Fenamic-acid family (2-arylaminobenzoic) identified as new scaffold direction; tetrazole bioisostere competitive with COOH; biphenyl > naphthalene > benzene in size effect. 0 GREEN / 0 YELLOW / 30 RED. Reframe for Misha stack: at η=0.5 max f_PC=0.50 asymptote; to cross f_PC=0.30 (MILD-MODERATE rescue threshold) need ΔG ≈ -7.06 kcal/mol (0.88 below current best). Routes: Phase 3c v3 ZINC22 bioactives 20k library (+0.5-1.0 kcal/mol expected), Phase 3c v4 fragment-growing on niflumic core (+1.0-1.5), or Phase 6b reversible covalent Lys-NH₃⁺ warhead (+2-3). A held, all scores unchanged. h01 reframed from “NORMAL monotherapy path” to “MILD-MODERATE adjunct lever with conditional NORMAL”; h03 S-tier primary lever unchanged. → STRC h01 Phase 3c v2 Expanded Screen 2026-04-23
#1 Pharmacochaperone Phase 5c: GREEN-LIGHT on K1141 site stability. Custom grid-based cavity analysis + MD stability on Phase 5a 20-snapshot trajectory (fpocket 4.0 brew build broken on qhull Delaunay; pivoted to numpy/scipy grid + Kabsch alignment). K1141 pocket Cα RMSF 0.62 Å vs global 1.23 Å — pocket residues 2× more rigid than average; local void volume 719-850 Å³ across 20 frames (15% CV) — pocket does not collapse or open into alternative geometry. Global cavity scan (snap_010, 26-direction burial ≥16/26, voxels ≥70 Å³): no alt cavity > 152 Å³ anywhere on the protein; nearest alt 18.7 Å from K1141 at 91 Å³. Phase 5b RED-LIGHT interpretation nailed: chemistry-limited, not site-limited. Phase 3c v2 expanded virtual screen targeted at K1141 with ensemble receptor docking confirmed as correct next move. A held, all scores unchanged. → STRC h01 Phase 5c Cryptic Pocket Analysis 2026-04-23
#1 Pharmacochaperone Phase 5: RED-LIGHT from ensemble re-docking. Phase 5a OpenMM MD pipeline validated on local Mac (62 ns/day OpenCL, 2 ns production in 45 min, 20 snapshots). Phase 5b Vina re-docking of Phase 4b top-5 leads + diflunisal against 20 MD-relaxed snapshots → all f_PC at [L]=10 μM < 0.10 (best: diflunisal positive control 0.083; best lead naphthalene-2-COOH 0.047). Phase 4b single-structure rankings over-optimistic by +0.36 to +0.92 kcal/mol — ensemble averaging revealed K1141 pocket less accommodating of carboxylate ligands than static snapshot. None of six cleared NORMAL-rescue threshold f_PC ≥ 0.50. A held, all scores unchanged (hypothesis mechanism intact; only current shortlist insufficient). Next step reordered: Phase 5c HEK293 wet-lab → Phase 3c v2 expanded virtual screen (DrugBank FDA + ZINC22 carboxylate tranche + fragment-based, ensemble filter ΔG ≤ −7.5 kcal/mol). h03 primary Misha lever reinforced (MILD rescue via AAV remains short-term path). → STRC h01 Phase 5 MD Ensemble Rescoring 2026-04-23
Misha compound-het stack integration: ref-entry added (Misha Compound-Het Therapy Stack Model) integrating #1 + #3 for Misha’s specific genotype (paternal 98 kb del + maternal E1659A). h03 AAV monotherapy physically cannot reach NORMAL hearing (≤ 25 dB) — ceiling-capped at MILD/MODERATE because non-transduced OHCs inherit 0 paternal + 0.5 × f_mat < θ from maternal, functional fraction capped at ε ≤ 0.5 → max 50% functional OHCs → ABR ~38 dB. h01 PC monotherapy CAN reach NORMAL at f_PC ≥ 0.50-0.75 (mild/moderate/severe E1659A); stack unlocks low-burden NORMAL at mild (9 combinations ε≤0.3 × f_PC≤0.5). Conclusion robust across θ ∈ {0.25, 0.35, 0.45}. Reframes h01 as the only monotherapy route to full cure for Misha; h03 remains S-tier as primary “meaningful rescue” path; clinical plan = parallel stack. Phase 5 MD on h01 priority strengthened (f_PC is the critical-path parameter). #1 + #3 scores unchanged; this is strategic framing. → Misha Compound-Het Therapy Stack Model
Misha compound-het stack integration: ref-entry added (Misha Compound-Het Therapy Stack Model) integrating #1 + #3 for Misha’s specific genotype (paternal 98 kb del + maternal E1659A). h03 AAV monotherapy physically cannot reach NORMAL hearing (≤ 25 dB) — ceiling-capped at MILD/MODERATE because non-transduced OHCs inherit 0 paternal + 0.5 × f_mat < θ from maternal, functional fraction capped at ε ≤ 0.5 → max 50% functional OHCs → ABR ~38 dB. h01 PC monotherapy CAN reach NORMAL at f_PC ≥ 0.50-0.75 (mild/moderate/severe E1659A); stack unlocks low-burden NORMAL at mild (9 combinations ε≤0.3 × f_PC≤0.5). Conclusion robust across θ ∈ {0.25, 0.35, 0.45}. Reframes h01 as the only monotherapy route to full cure for Misha; h03 remains S-tier as primary “meaningful rescue” path; clinical plan = parallel stack. Phase 5 MD on h01 priority strengthened (f_PC is the critical-path parameter). #1 + #3 scores unchanged; this is strategic framing. → Misha Compound-Het Therapy Stack Model
#9 Hydrogel: A held, Mech 4 held, Deliv 3 held. Phase 4i Kd × Kd 5×5 sensitivity sweep on therapeutic window: f ≥ 0.3 reachable on 80% of plausible grid (PASS 56% single-dose ≤ 5 mg + MARGINAL 24% at 5-20 mg); FAIL 20% only at WH2 Kd ≥ 1 mM (Tβ4 analog floor) where G-actin toxicity ceiling hits before f=0.3 achievable. Strategic reorder: WH2×F-actin Kd dominates, TMEM145 Kd secondary. Phase 2c WH2 bundling (HEK293 GFP-actin confocal) now wet-lab gate #1; TMEM145 SPR/BLI → #2 (ipSAE already places in operable band). Verdict A_hold_multi_dose_ok — S promotion still requires WH2 Kd narrowing. → STRC h09 Phase 4i Kd Sensitivity Sweep 2026-04-23
#1 Pharmacochaperone post-audit housekeeping: A held. TPSA docstring corrected in phase3b_virtual_screen.py (descriptor-only, not scoring; STRC extracellular so CNS TPSA bracket irrelevant; no fabricated replacement range). Druggability cross-phase incomparability flagged in Phase 1/2/2b. → STRC h01 Parameter Provenance Audit 2026-04-23

2026-04-22

#1 Pharmacochaperone: S → A. Mech 4→3. STRC extracellular; 4PBA repurposing branch killed. → STRC Repurposed FDA Chaperone Branch
#1 Pharmacochaperone Phase 4f SMOKE: numerically meaningless (ΔΔG 20,439 kcal/mol). Method error; Phase 4f production canceled. → STRC Pharmacochaperone Phase 4f Interface Rescue SMOKE
#1 Pharmacochaperone Phase 4g: 4PBA first mut-preferring compound (−0.32 mut-wt). Opens repurposing branch (later killed). → STRC Pharmacochaperone Phase 4g Repurpose Screen

2026-04-21

Note created. Initial assignment 24 hypotheses. S: {#1 Pharmacochaperone, #2 Piezo, #3 Mini-STRC, #4 Strategy B}. Killed: {OTOA chimera, ZP prion, protein replacement}. Paused: {Sonogenetics, Recombinases}.
#1 Pharmacochaperone Phase 4a: PASS 4/5. K1141 pocket reproduces across CIFs. → STRC Pharmacochaperone Phase 4a Pocket Reproducibility
#1 Pharmacochaperone Phase 4b: PASS on LE. Leads beat diflunisal 29-57%. → STRC Pharmacochaperone Phase 4b Smoke Test
#1 Pharmacochaperone Phase 4c: FAIL. All 5 leads prefer WT over E1659A (mean −0.455 kcal/mol). Aspect (a) flagged. → STRC Pharmacochaperone Phase 4c WT Decoy
#1 Pharmacochaperone Phase 4e: soft-FAIL (margin 0.15 vs 0.20). Scorer limitation; re-opens as Phase 4b sub-test. → STRC Pharmacochaperone Phase 4e Off-Target Selectivity

Connections

[part-of] h01 hub

STRC Research

Explorer

log

h01 log

2026-04-27 (wetlab handoff packet v0.1 delivered)

2026-04-27 (14:38 — Phase 9b min-relax multi-frame MM-GBSA smoke delivered)

2026-04-27 (14:13 — Phase 9b repaired matched Amber inputs + minimized endpoint smoke delivered)

2026-04-27 (13:31 — Phase 9b geometry sanity delivered: matched complexes are physically clashed)

2026-04-27 (13:12 — Phase 9b matched Amber inputs + one-frame MMPBSA.py read delivered)

2026-04-27 (11:54 — Phase 9b MM-PBSA tooling preflight delivered: software ready, matched inputs blocked)

2026-04-27 (11:29 — Phase 9b-lite residue-field decomposition smoke delivered: broad E1659A>WT electrostatic sign survives APBS-free post-processing)

2026-04-27 (00:30 — Phase 9 orthogonal cross-checks plan opened: 5 SKELETON phase notes for FEP/TI, holo MD+MM-PBSA, umbrella PMF, QM/MM, blind neural docking)

Ranking delta

Artifacts

2026-04-26 (21:37 — Phase 5q n=20 expansion DELIVERED: σ_within tightened from 1.45× → 1.36× mean / 1.17× median, rank-2 flips adamantyl-Cl > 1-indanyl-Cl, all three τ-tied within SEM)

2026-04-26 (20:30 — paper-ready batch: Phase 8c v5.3 ADMET + Phase 5e v5.3 mut-ensemble + Figure 4 + paper outline §4.6 + lead committee matrix)

2026-04-26 (20:00 — Phase 5q STRC v5.3 within-target τRAMD: σ_within = 1.45× → cross-target bound ≤ 2.9× ≪ 5×, v5.3 kinetic-selectivity rescue falsified pre-emptively, TRPM4 v5.3 τRAMD SKIPPED)

2026-04-26 (17:xx — h01 state audit + TRPM4 closure reconciliation)

2026-04-26 (13:1x — Phase 5q v5.3 acyl-sulfonamide top-3: Boltz-2 + Vina-v3 rigid AGREE on rank-1 = 1-indanyl_SO2Me_-Cl ipTM 0.645)

2026-04-26 (10:5x — Phase 5e-v2 v5.2-shortlist: Vina vs τRAMD ρ = −1.000 anti-correlation, Gasteiger zero-anion artifact identified as load-bearing — Vina ΔG axis demoted for anion-leads until Meeko/RESP upgrade)

2026-04-26 (10:4x — Phase 5m production τRAMD: 25/25 UNBOUND, CONHOH confirmed shortest, within-STRC spread 1.7× falsifies kinetic-selectivity rescue, v5.3 pipeline ACTIVATED)

Ranking delta

Artifacts

2026-04-26 (09:4x — parallel session while τRAMD production runs: calnexin retention NEGATIVE, v5.3 acyl-sulfonamide library 6/12 PASS, sci-hub 4 ERAD-timing PDFs landed)

(a) Phase 5n-lite — N-glyc map + calnexin retention proxy on STRC

(b) Phase 5p — v5.3 acyl-sulfonamide library design + triage

(c) Sci-hub ERAD-timing literature — 4 PDFs

Production status (in progress)

Ranking delta

Artifacts

2026-04-26 (09:0x — Phase 5l + 5m smoke: AF3 fragment ladder selected, τRAMD pipeline validated, lead unbinds at 17.9 ps biased on 151-aa fragment)

(a) Phase 5l — AF3 K1141 fragment ladder + selection gate

(b) Phase 5m smoke — τRAMD pipeline end-to-end validation

Pipeline lessons captured

Ranking delta

Artifacts

2026-04-26 (08:3x — Phase 8g-v3-lite + 8h-lite #9: gate diagnostic + dose-escalation rescue. Heavy-queue replanned)

(a) Phase 8g-v3-lite — pocket-max-φ resample

(b) Phase 8h-lite #9 — PBPK dose-escalation sweep

Ranking delta

Artifacts

2026-04-26 (01:5x — DR1 next_step delivered: PBPK ODE + interface map + lit-gap-map. deliv-3 ceiling reinforced from delivery side)

(a) Phase 8h-lite #7 — Cochlear-compartment PBPK ODE

(b) Phase 8h-lite #8 — K1141 vs partner-interface map

(c) STRC E1659A QC rate-limiting step — lit-mining gap-map

Summary numerics

Ranking delta

2026-04-26 (01:3x — DR1 distillation: 5 clinical-precedent atomic notes; rate-limit framing + PBPK gap surfaced)

Ranking delta

2026-04-26 (02:0x — DR4 v2 prose draft + Stream A/B duplicate-catch + scripts inventory cleanup)

2026-04-26 (early-AM — Phase 8h-lite #6 Tier-1 refresh; phosphonate prodrug flag raised)

2026-04-26 (01:30 — DR2 distillation + competitive re-canvass: 3 new atomic notes; landscape extended +3 entries; white-space CONFIRMED)

Ranking delta

2026-04-26 (01:5x — Phase 7I follow-on triple delivery: ADMET re-check + pose ensemble stability + APBS off-target on Tier-1)

2026-04-26 (01:3x — Phase 7I v52 combined analysis: v52_top3 closed; lead adamantyl_CONHOH_-Cl is a Boltz-2 tie)

2026-04-26 (01:2x — Phase 7H v52_rest11 Boltz-2 analysis DELIVERED; ZBG-pivot evidence for v5.3)

2026-04-26 (01:2x — DR3 distillation + improvement plan)

Ranking delta

2026-04-26 (01:1x — DR4 review + enhancement plan)

2026-04-26 (01:1x — DR5 intratympanic delivery deep research ingested + cleaned)

2026-04-26 (post-midnight — Phase 8h-lite DELIVERED; deliv 3 → 4)

2026-04-25 (night — literature ingest: Schneider Ed. 2014 De novo Molecular Design)

2026-04-25 (night — literature ingest: Chipot & Pohorille 2007)

2026-04-25 (night — Phase 7H DELIVERED via Boltz-2 local; mech 4 → 4−)

2026-04-25 (evening — Phase 7H BLOCKED, redirect to Boltz-2 local)

2026-04-25 (pre-dawn — Phase 8g off-target panel + Vina limitation documented)

2026-04-25 (overnight — Phase 5k-WT matched ensemble + Phase 8d/8e v5.2 running)

2026-04-24 (late night — Phase 8c v5 ADMET triage + overnight queue setup)

2026-04-24 (late night — Phase 8 v5 library Coulomb-aware design)

2026-04-24 (night — Phase 5k ensemble APBS, Tier-0 PASS)

2026-04-24 (night — Phase 4i APBS pose-transplant, Tier-4)

2026-04-24 (evening — Phase 5j APBS + Phase 5c-mutant)

2026-04-24 (evening — Phase 6c hERG extension)

2026-04-24 (evening — Phase 7H AF3 validation queued)

2026-04-24 (evening — Phase 7 ADMET-AI triage)

2026-04-24 (evening — v4 delivery + Phase 4c-v3b + Phase 4d)

2026-04-24 (afternoon — Phase 6c full 6-target panel)

2026-04-24 (afternoon — Phase 6c COX selectivity)

2026-04-24 (afternoon launch — Phase 3c v4)

2026-04-24 (afternoon delivery)

2026-04-26 (13:1x — Phase 5q v5.3 acyl-sulfonamide top-3: Boltz-2 + Vina-v3 rigid AGREE on rank-1 = `1-indanyl_SO2Me_-Cl` ipTM 0.645)

2026-04-26 (01:3x — Phase 7I v52 combined analysis: v52_top3 closed; lead `adamantyl_CONHOH_-Cl` is a Boltz-2 tie)