Stored resonance strength gave no clear connection prediction gain in selected Allen mouse cortical slices

We tested whether stored resonance strength adds information for detected chemical connections after bandwidth, passive cell properties, cell labels and recording controls. We used Allen Synaptic Physiology r2.1 data and five folds that held out whole inferred donors. The main mouse visual cortex sample had 2410 tested directions, 236 detected connections and 197 donors. Adding two log2 resonance ratio terms changed log loss by -0.116% (positive means less error). The conditional 95% donor bootstrap interval was -0.585% to 0.393%. A separate human temporal cortex fit gave 1.122% with interval -0.788% to 2.965%. This test did not resolve added prediction value in the main sample. Most stored ratios were close to one. Missing physiology, a limited frequency window, slice selection, inferred donor keys, and fixed-prediction intervals limit the result. The package preserves code, candidate measurements, selected pairs, predictions and source hashes.

Allen Institute · resonance · chemical connectivity · donor validation · null result