Thirty targeted one-contact changes remove a fitted response similarity slope in 8,128 connected MICrONS pairs

We tested the stability of a reported link between movie response similarity and multiple synaptic contacts in one MICrONS mouse. We kept 8,128 connected cell pairs, their predictors, and the linear model fixed. The model controlled for both cells and proximity lengths. Its slope was 2.67 percentage points per standard deviation of response similarity. An exact sort of regression weights showed that 30 selected one-contact changes made the slope nonpositive. Every allowed set of 29 changes left it positive. The changes could add one contact to a single-contact pair or remove one from a two-contact pair. With at most one change per source cell, 31 changes were needed. Additions alone needed 34 changes; deletions alone needed 42. Uniform random changes left the slope positive in all 10,000 trials at each of 30, 60 and 150 changes. This is a fixed-design sensitivity result. It does not estimate an error rate or show that any recorded contact is wrong. One mouse, selected connected pairs, fixed geometry and response estimates, and the linear model limit the result. Code, selected data, correction sets and separate regression checks support reproduction.

MICrONS · synaptic contacts · response similarity · measurement error · sensitivity analysis