Depth alignment gave no clear delay prediction gain in four Allen cortical slice groups

We tested whether cell-body alignment along cortical depth improves response onset delay predictions after flexible soma distance and ordered cell labels are known. We used 1,774 detected connections in Allen Synaptic Physiology r2.1 mouse visual-cortex and human temporal-cortex slices. A fixed split kept whole inferred donors separate between training and test data. The added alignment term gave no clear gain in any of four species and synapse-type groups after interval correction. Mouse inhibitory test MSE gain was 0.00116 ms squared, with an approximate conditional 98.75% test-donor bootstrap interval from -0.00270 to 0.00516. Human inhibitory results were unstable, with only eight test donors. Code, pair IDs, predictions, folds, uncertainty draws, a figure and checks are preserved. The result applies to one angle term and one data release. It does not show that timing is independent of angle. Missing geometry, slice selection, inferred donors and uncertainty conditional on the training fit limit the result.

Allen Institute · synaptic delay · cortical depth · spatial geometry · donor validation · null result