Measured movie similarity gave no clear added wiring gain over model similarity with test neurons excluded in one MICrONS mouse

We tested whether measured movie response similarity adds wiring information beyond a released model response summary. We used MICrONS functional connectomics v1 from one mouse. The panel had 295,371 candidate directions with positive axon-dendrite proximity and 8,128 recorded connections. Five outer groups excluded both test neurons from each wiring fit. Inner groups chose the rule and its penalty. The main test had 58,595 directions and 1,608 connections. Adding measured similarity reduced log loss from 0.113126788 to 0.113118423, a 0.00739% gain. A conditional range from neuron weights was -0.05492% to +0.07099%. Four of five fits selected the rule with model similarity alone. Model similarity reduced control loss by 0.26442%; measured similarity alone reduced it by 0.15145%. A second split, whole-cell sources and a wider grid gave gains from -0.00897% to +0.02348%. Their conditional ranges crossed zero. These results give no clear added gain from measured similarity in the main test. They do not establish equal information or the absence of a biological effect. The summaries use different movie sets. The ranges fix predictions and omit training and animal variation. We preserve code, selected data, source hashes, predictions, all checks, figures and repeat instructions.

functional connectomics · MICrONS · visual cortex · signal correlation · digital twin · neuron exclusion · prediction