Cell identity gave no clear added gain in a ridge test of late 50 Hz response changes in 1,126 mouse cortical connections

A connection map gives a source and a target. We tested whether other connections of the same neuron help predict its late response change during a 50 Hz train. We used 1,126 detected connections from two mouse cortical projects in Allen Synaptic Physiology r2.1. The scalar outcome was the median normalized change from pulse 1 to pulses 6–8. The comparison ridge model used cell class, strength, sampling, geometry and experiment identity. Nested pair folds kept reverse directions together. Source and target identity had separate penalties and separate terms for each synapse sign. Adding both cell roles raised mean squared test error from 0.07360819 to 0.07420878, an increase of 0.82%. Five fixed pair splits gave error changes from a 0.08% decrease to a 2.87% increase. Source identity alone gave a main error decrease of 0.073%, with a fixed-prediction donor range from a 0.32% increase to a 0.50% decrease. Only 466 test entries had another same-sign input or output of the same cell in training. Their joint gain was also uncertain. A stricter repeat-count rule and omission of amplitude terms gave no joint gain. The method recovered large synthetic cell signals. An independent dense calculation agreed with saved predictions, and a full numerical repeat matched 58 result files. This extends earlier initial-change and smaller late-change tests. It is a completion result for sparse slice recordings, not a test of biological independence or new-animal prediction. The package preserves code, source hashes, row IDs, fitted states, results, figures and repeat instructions. Original source data remain external.

short-term synaptic plasticity · cell identity · mouse cortex · measurement completion · paired recordings · ridge regression · source and target neurons · Allen Synaptic Physiology