Eight active cell measures gave no clear gain in a logistic chemical connection test of 12,432 mouse V1 directions

We tested whether eight active cell measurements help predict detected chemical connections. The controls included cell labels, geometry, recording data and passive properties. The source was Allen Synaptic Physiology r2.1 small. The main sample had 12,432 mouse V1 directions, 1,177 detected connections and 668 inferred donors. Nested logistic ridge tests kept donors separate. All models included measurement coverage controls. Separate source and target active values changed mean test log loss from 0.25707578 to 0.25823471, a 0.451% increase. Eight common cell matching terms then raised loss to 0.25893874, a further 0.273% increase. The 95% donor resample ranges for error reduction were -1.349% to +0.397% and -0.448% to -0.086%, with predictions fixed. The same procedure recovered an artificial matching signal, with 7.83% less log loss. Planned checks and an added source range check are preserved. Source selection, missing connections, inferred donors, measurement error and the fixed model family limit inference. The result limits a prediction rule; it does not show that intrinsic physiology has no role in circuit function. The package gives code, source hashes, plans, fitted models, test predictions, results, figures and repeat steps. Original Allen measurements remain external.

intrinsic physiology · chemical connectivity · mouse visual cortex · cell matching · grouped validation · missing measurements · Allen Synaptic Physiology