With surface opportunity fixed, removal of 1.16–17.35% of contact counts moved predicted direction by 45 degrees in 27 mouse retinal cells

We tested how contact removal changes a direction predicted from anatomy in one mouse retina. The main set had all 27 matched On-Off direction-selective ganglion cells in four published types. We pooled starburst contact counts in eight angular bins and divided them by local surface opportunity. We kept opportunity fixed and allowed continuous removal from each bin. An exact rule gave the smallest removal that could change a model direction by 45 degrees. These limits were 1.16–17.35% of total contact counts, with a median of 10.31%. At a 10% total-count budget, 15 of 27 predictions had a guaranteed change of less than 45 degrees. With removal capped at 10% in every bin, 26 of 27 had that guarantee. Sixteen and thirty-six bins gave median total-count limits of 10.14% and 9.81%. Independent linear programs checked all 54 primary boundaries and both formulas on 400 additional boundaries. The package provides code, scalar results, exact removal witnesses, source hashes and repeat instructions. Contacts include incidental membrane contacts. The source arrays pool partner identities. These are sensitivity limits for a fixed contact predictor. They do not estimate reconstruction error rates or physiological effects of cell loss.

retina · direction selectivity · starburst amacrine cells · structure and function · contact surface · sensitivity analysis · connectomics · reproducible methods