Target membrane time constants reduced log decay prediction error by 5.1% in a fixed ridge test of 650 mouse cortical connections

We tested whether a target cell membrane time constant helps predict the decay of a measured synaptic voltage response. We used the Allen Synaptic Physiology r2.1 small database. The main panel had 650 mouse V1 connections from 233 inferred donor groups. A nested ridge test excluded whole donor groups from training. The baseline used current kinetics, cell labels, and distance. Adding target membrane time reduced squared log decay error from 1.30508 to 1.23884, or 5.08%. A 10,000-draw donor range with predictions fixed was 0.47% to 9.45%. A later 1,000-draw check that refit the models gave a range from 0.0011% to 9.35%; the lower endpoint was effectively zero at this precision. On the same 644 connections, the gain fell from 4.73% to 0.86% after input resistance adjustment. The refitting range for the adjusted gain was -1.11% to 2.80%. A secondary human panel of 172 connections gave a gain of -0.04%. Removing fits near default limits gave an uncertain mouse gain. All selected stored kinetic fields matched accepted component-fit aggregation exactly. These results concern selected local recording graphs and a fixed model family. They do not establish a general membrane mechanism or a species effect. The package preserves original code, source hashes, selection IDs, coefficients, predictions, loss tables, bootstrap draws, figures, and repeat instructions. Original source measurements remain external.

synaptic physiology · membrane time constant · temporal integration · cortical circuits · structure and function · group validation · ridge regression · reproducible research