STRC Gamma-Poisson Transduction Model

Context

The STRC Dual-Vector vs Single-Vector Transduction note contains v1 model numbers (single 67.4%, dual 1.2%, ratio 56.5x). These were SUPERSEDED on 2026-03-19 by a more accurate Gamma-Poisson (negative binomial) model.

v1 numbers are WRONG — the receptor competition correction was double-counted.

The Problem with Simple Poisson

v1 model: simple Poisson distribution for virus uptake. Issue: assumes every cell in the cochlea has identical probability of being transduced. In reality, cells vary enormously in their accessibility to the viral vector (depth from injection site, aqueous flow, extracellular matrix density). This cell-to-cell heterogeneity violates the Poisson assumption.

Gamma-Poisson (Negative Binomial) Model

The correct model: viral uptake at each cell follows a Gamma-Poisson (negative binomial) distribution:

  • λ: mean particles per cell
  • k: overdispersion parameter (k → ∞ = pure Poisson, small k = high heterogeneity)

Calibration from Experimental Data

Two data points from Omichi et al. (2020):

  • Single AAV2 via RWM: 83.9% OHC transduction (mouse)
  • Dual AAV2 via RWM: 65.6% OHC transduction (mouse)

Fitted parameters:

  • λ_mouse = 8.11 (mean particles per OHC)
  • k = 0.734 (overdispersion = 11.1x — significant heterogeneity)

Human Scaling

Mouse → human scaling factors:

  • Perilymph volume: mouse 2.5 µL → human 191 µL (76x)
  • Injection fill ratio adjustment
  • Net scaling: 0.613x

λ_human = 4.97, k_human = 0.734

New Predictions (Clinical Titer 3.75×10¹² GC/mL)

ConstructOHC TransductionOHC Count
Single-vector mini-STRC77.8%9,336 of 12,000
Dual-vector (R=30% recombination)16.4%1,967
Dual-vector (R=50% recombination)27.3%3,278
Dual-vector (R=100% recombination, theoretical max)38.2%4,589

The 56.5x ratio from v1 was too extreme (double-counted receptor competition). The correct advantage is 2.8-4.7x depending on recombination efficiency. Still decisive — even best-case dual vector is 2x worse than single.

Why v1 Was Wrong

v1 explicitly penalized dual-vector for receptor competition (AAV-A and AAV-B fighting for the same surface receptors). But the calibration data itself (Omichi dual = 65.6%) ALREADY encodes that competition. Double-counting produced an artifactually extreme ratio.

The Gamma-Poisson v2 model calibrates directly to both data points simultaneously, so receptor competition is implicit in the fitted parameters.

Updated Site

Connections