Shared neighbors gave no clear prediction gain for response size in nested ridge tests of 903 mouse V1 connections

We tested whether shared neighbors help predict measured synaptic response size. We used Allen Synaptic Physiology r2.1 data from adult mouse primary visual cortex. The voltage sample had 903 connections, 319 mouse IDs, and 430 recordings. Both focal directions and all four third-cell directions needed valid tests. The baseline used cell labels, geometry, recording information, target intrinsic properties, reverse connection status, and local connection counts. We added a shared-neighbor fraction with separate excitatory and inhibitory slopes. Nested ridge tests kept every test animal out of fitting and penalty choice. Mean squared log voltage error increased by 0.47%. The 95% range from 199 complete animal resamples and refits was -3.27% to +1.66% error decrease for one split. Six voltage fits below 1 microvolt gave 59.1% of baseline log error. An asinh score stayed finite at zero and gave a 0.74% voltage error increase. Its complete-refit range was -5.07% to +1.38%. The asinh current test gave a 1.19% increase in 784 connections. Threshold, source-class, count, and presence checks gave no clear overall gain. These results limit this graph feature and linear prediction rule in selected slice recordings. They do not exclude circuit assemblies or causal effects of clustering. The package has code, source URLs and hashes, model states, predictions, resample results, figures, and repeat instructions. Original source data stay external under Allen Institute terms.

connectivity · synaptic strength · common neighbors · mouse visual cortex · multipatch recordings · cross validation · measurement sensitivity