A membrane time constant matching term gave no clear added gain in a logistic test of 8,667 mouse cortical directions

We tested whether similar membrane time constants help predict a detected chemical connection. We used Allen Synaptic Physiology r2.1 small. The main sample had 8,667 tested directions, 980 detected connections and 619 inferred mouse donors. A logistic reference model used cell labels, geometry, sampling, input resistance and separate nonlinear effects of each cell's time. The added term was their absolute log2 time difference. Nested validation kept each donor in one fold. The added term changed mean log loss from 0.29116681 to 0.29116463, a 0.00075% reduction. The 95% donor resample range was -0.130% to +0.126%, with predictions fixed. A later model with separate matching terms for four broad class pairs gave a 0.023% gain, with range -0.369% to +0.400%. Among inhibitory pairs, raw detection fell from 18.5% to 7.3% across bins of time difference. Reference predictions reproduced much of that fall. Planned checks gave no clear added gain. The method recovered a large artificial matching signal. These are selected slice recording graphs. Unmeasured response noise, missed connections, cell selection, inferred donors and measurement error limit inference. The result limits a prediction rule; it does not show that membrane timing has no biological role. The package preserves code, source hashes, cohort IDs, model states, results, figures and complete repeat instructions. Source measurements remain external.

connectivity · membrane time constant · patch clamp · prediction · donor validation · cell physiology