Resting variation gave no clear late voltage prediction gain after strength controls in ridge tests of 993 mouse cortical connections

Connection models need rules for signal size during repeated activity. We tested whether resting response variation adds late voltage prediction information. Baseline inputs included cell class, geometry, recording conditions and response strength. We used separate scalar summaries from Allen Synaptic Physiology r2.1. Ridge models used five outer donor folds and four inner donor folds. Each donor had equal weight. The primary sample had 993 connections from 433 inferred mice. Adding resting variation reduced error by 23.73% when strength features were omitted. With strength included, added variation increased error by 0.75%. The central 95% donor resample range for error reduction was -1.62% to +0.10%. These ranges keep predictions fixed and omit uncertainty from new model fits. Wider penalty choices gave an uncertain +0.51% gain, with a range of -2.20% to +3.38%. An added second-response control reduced error by 20.74% in 986 connections. Human results were uncertain. The large gain without strength controls did not persist clearly after those controls in the tested models. Summaries can share trials and fitting assumptions. This retrospective prediction test does not estimate release probability or release-site count. The package preserves code, provenance, predictions, all checks, figures and repeat instructions. The exact source database remains external under its original terms.

cortical connectivity · synaptic variability · short-term plasticity · response strength · donor validation · ridge regression