Soma distance reduced squared log delay test error by 3.61% beyond cell class and recording controls in 2,242 mouse cortical connections

We tested whether distance between cell bodies improves synaptic delay prediction beyond cell class and recording controls. We used 2,242 detected chemical connections from three mouse cortical projects in Allen Synaptic Physiology r2.1. All measurements from a test donor were excluded from fitting and model selection. Adding distance to ridge regression reduced mean squared log delay error from 0.158723 to 0.152989, or 3.61%. The central 95% donor resampling range was 2.09% to 5.17%. This range keeps test predictions fixed. All five test folds improved. Five donor partitions gave decreases of 3.09% to 3.64%. Separate clamp modes, known cell classes, shorter distances, and removal of boundary fits kept positive gains. Later project subsets showed no clear gain in two of three projects. All 2,242 distances matched values rebuilt from cell coordinates. All 2,665 reported source delays matched accepted component fit averages. These data support a small distance benefit in this prediction task. They do not give axon conduction speed or establish transfer to another laboratory. The source contains sampled slice graphs and incomplete cell labels. The package preserves original code, source hashes, cohort IDs, folds, model states, donor error totals, figures, and repeat instructions. Original Allen records stay external under their source terms.

synaptic delay · soma distance · mouse cortex · structure and function · paired recordings · ridge regression · Allen Synaptic Physiology