Matching three 1–15 Hz impedance summaries gave no clear connection prediction gain in a ridge test of 3,315 mouse cortical directions

We tested whether local chemical connections link cortical cells with similar frequency responses. We used Allen Synaptic Physiology r2.1 and three impedance summaries from 1 to 15 Hz. The main sample had 3,315 tested directions, 288 detected responses, and 217 inferred mouse donors. A ridge logistic model used cell labels, geometry, recording controls, passive properties, and separate nonlinear effects of each cell’s frequency features. Three absolute log2 differences were then added. Nested validation kept each donor out of model training and penalty selection. Mean test log loss changed from 0.263846 to 0.263729 nats per direction. The gain was 0.044%, with a descriptive 95% donor resample range from -1.23% to +1.40%. The predictions were fixed during resampling. Only 60 of 1,115 selected mouse cells had a peak ratio of at least 1.1. Many peaks and 3 dB frequencies were near the analysis limits. In a separate fit of 1,241 human directions, matching reduced error by 0.60% relative to the endpoint model, but its point error remained above the simpler passive and recording model. The mouse test gives no clear support for the specified matching rule. It does not show that resonance has no role in circuits. The package gives original code, an internal plan, source hashes, record IDs, folds, model states, predictions, figures, and repeat instructions. Source measurements remain external.

functional connectivity · membrane impedance · resonance · cortex · frequency selectivity · donor validation · local circuits