STRC Research

STRC h06 Parameter Provenance Audit 2026-04-23

4 min read

STRC h06 mRNA Therapy Parameter Provenance Audit 2026-04-23

Post-fix to the Batch 2 audit findings (STRC Cross-Hypothesis Parameter Audit 2026-04-23 Batch 2). Replaces ”⚠ UNSOURCED” on two load-bearing protein half-life constants with lit-anchored citations while keeping the numbers explicit as order-of-magnitude estimates. Hill constants remain CIRCULAR FIT — no primary-lit OHC splicing dose-response exists to back them.

Files touched

  • models/mrna_lnp_pkpd_integration.py
  • models/mrna_lnp_pkpd_integration_v2_ohc_tropism.py

Lit-anchoring applied

STRC_HL_D = 14 days

Previously: ”⚠ UNSOURCED — no stereocilin t½ measurement”.
Now framed as lit-anchored order-of-magnitude estimate within the stereocilia slow-turnover regime.

Primary anchors:

  • Zhang et al. 2012 Nature 481:520–524. “Multi-isotope imaging mass spectrometry reveals slow protein turnover in hair-cell stereocilia.” PMID 22246323, DOI 10.1038/nature10745. Direct quote from abstract: “the stereocilia were remarkably stable, incorporating newly synthesized protein at <10% per day” → bulk stereocilia t½ > ~7 d. Stereocilia proteins remained stable in pulse-chase up to 150 days.
  • Mauriac & Barr-Gillespie 2024 PNAS 121:e2405455121. “Proteins required for stereocilia elongation during mammalian hair cell development ensure precise and steady heights during adult life.” PMID 39320919, DOI 10.1073/pnas.2405455121. GPSM2 t½ 9–10 days at stereocilia tips — the closest specific lit measurement to what STRC turnover might be.

Honest caveat in scripts: STRC-specific t½ was never measured. 14 d is within the lit-supported stereocilia regime but could span 7 to 150+ days per Zhang 2012 range.

RBM24_HL_D = 2 days

Previously: ”⚠ UNSOURCED — no OHC-RBM24 t½ measurement”.
Now framed as lit-anchored median-proteome estimate, not RBM24-specific.

Primary anchor:

  • Schwanhäusser et al. 2011 Nature 473:337–342. “Global quantification of mammalian gene expression control.” PMID 21593866, DOI 10.1038/nature10098. Median protein t½ = 46 h ≈ 1.9 d in mouse NIH 3T3, measured by SILAC + LC-MS/MS. BioNumbers ID 106377.

Honest caveat in scripts: OHC-specific RBM24 t½ never measured; 2 d is proteome-median plausible.

Hill-coupling constants — no fix possible

HILL_K = 200, HILL_N = 2, MAX_BOOST = 3.0 remain ⚠ CIRCULAR FIT. No primary-lit OHC RBM24→STRC splicing dose-response experiment exists. 2026-04-17-sun-rbm24-strc-splicing establishes qualitative splicing control but does not publish a dose-response curve.

Model implication: the ODE cannot fail by construction — values were set so “therapeutic” outputs emerge at “realistic” doses. A sensitivity sweep (HILL_K ∈ [50, 500], MAX_BOOST ∈ [1.5, 5]) would bound what the model actually predicts across parameter uncertainty; not yet run. Flagged as a follow-up gate.

LNP tropism constants — partial fix

  • LNP_ENDO_ESCAPE = 0.02 — stays ⚠ “within <10% published range, specific cite missing” — Patel 2019 PNAS candidate anchor exists but was not retrieved in this audit.
  • LNP_UNTARGETED = 0.008 (≈ 96/12,000 OHC) — stays ⚠ unsourced specifically. Prior “Gao 2020 PMID 32493791” phantom identified as COVID-19 paper; real candidates Gao 2018 Nature (Beethoven lipid-RNP) and Yeh 2018 Nat Biomed Eng (CBE cochlear) are listed in the provenance block but not retrieved for exact numerical backing.
  • LNP_COCHLEAR_TROPIC = 0.05, LNP_OHC_TARGETED = 0.20 — remain correctly flagged as hypothetical next-gen / ligand-targeted values.
  • ER_CEILING_FOLD = 5.0 — remains ⚠ MODEL HEURISTIC (Sahin 2014 = wrong-paper-match). No primary-lit UPR fold threshold for cochlear mRNA exists. Documented in v2 script header.

Ranking delta

#6 mRNA Therapy: A held. Mech 3, Deliv 2, Misha-fit 2. No axis change — the lit-anchoring upgrades script defensibility but does not change the underlying mechanism/delivery/Misha-fit score. Downstream dependent script rbm24_mrna_dose_response.py generating script was not located in repo audit (JSON only); if recovered, apply same lit-anchoring to source.

#4 Strategy B mRNA-LNP: B held. Inherits h06 Hill risk (the generating JSON is shared). No independent phantoms in h04 scripts beyond batch 2 audit.

Still-open gates (not closed this round)

  1. HILL sensitivity sweep — bound ODE outputs over HILL_K × MAX_BOOST uncertainty grid. Independent, low-effort, reveals model robustness.
  2. Patel 2019 / Gao 2018 / Yeh 2018 retrieval — confirm LNP endo-escape and OHC efficiency primary cites.
  3. rbm24_mrna_dose_response.py — generating script not in repo. Its math would tell us whether the “hill_km = 200” value is an Michaelis-Menten fit to Sun 2026 or genuinely invented.

Connections

Graph View

Backlinks