A nested logistic test selected no local input offset term for motion direction in 1,820 MICrONS neurons with known axes

We tested whether input soma location predicts which of two opposite motion directions a neuron prefers. We used MICrONS v1507 from one mouse and supplied each target's orientation axis. We scanned all 303,164,967 connection rows. We kept 1,478,285 input edges from sources with a class and soma, including sources without functional matches. The main test used 1,820 pyramidal cells and local source somata within 200 micrometres. Controls included position, cell labels, recording, counts, input spread and the pooled input centroid. Five outer tests kept whole spatial cubes out of training. Three inner tests chose separate penalties for controls and the inhibitory minus excitatory centroid. All five main tests selected exact removal of the offset. Both rules gave Brier error 0.234148. A later forced-offset test raised error by 0.0554%. A map that varied with position also gave no gain. The local classified inputs supplied a median 7.08% of each target's recorded incoming synapses. This result limits the tested prediction rules. It does not show that spatial inhibition has no role in motion tuning. One animal, model-derived tuning and incomplete source coverage limit the test. We preserve code, source hashes, the selected input graph, features, predictions, model states, figures and repeat instructions.

connectome · MICrONS · motion direction · excitatory inputs · inhibitory inputs · soma location · spatial validation · negative result