Soma distance reduced squared delay prediction error by 10.9% in fixed ridge tests of 729 inhibitory connections from mouse V1

We tested whether soma distance adds transmission delay information beyond cell labels in sampled cortical circuits. We used Allen Synaptic Physiology r2.1. Five fixed ridge tests kept whole inferred donor groups out of training. In 729 mouse V1 inhibitory connections from 265 donor groups, adding distance reduced mean squared prediction error by 10.86%. Root mean squared error fell from 0.459 to 0.434 ms. A later donor bootstrap with model refits gave a central 95% gain range of 3.43% to 19.83%. The gain was 8.82% in 693 inhibitory connections at most 200 micrometres apart. The adjusted inhibitory slope was 0.325 ms per 100 micrometres, with a donor-cluster interval of 0.246 to 0.404. Positive slopes remained within source neurons and within target neurons. Checks using current-clamp delays, source Cre labels and response strength also kept positive inhibitory slopes. Prediction gains in 410 mouse and 116 human excitatory connections were less certain. All 2,665 nonmissing released latencies matched weighted values from accepted source fits. A later test on the common cohort of a related SCI study gave a 9.15% gain in 712 inhibitory connections, with a full-refit range of 2.82% to 17.62%. A fresh rebuild reproduced 92 numerical files exactly. The package preserves code, cohort IDs, predictions, resample draws, aggregate data, figures, provenance and repeat instructions. These are selected slice recording graphs. Soma distance omits axon and dendrite routes, and response onset includes release and filtering. The results support a distance term in the tested models. They do not identify conduction speed or a species effect.

synaptic delay · structure and function · mouse visual cortex · synaptic physiology · soma distance · donor validation · connectome models