Vertical squared distance gave no clear connection prediction gain in selected Allen human cortical slices

We tested whether vertical squared distance improves detected chemical connection prediction beyond a common radial distance term, cell labels, recorded cell depths, and test-spike count. Allen Synaptic Physiology r2.1 supplied 1,787 human directions, 160 detected connections, and 54 inferred donors. Five-fold donor validation gave a log loss gain of -0.000016 nats per direction, with a conditional 95% donor bootstrap interval from -0.000081 to 0.000039. A flexible radial model and four human subset checks gave no clear gain. In a mouse reference sample, a gain with a quadratic radius became uncertain with a radial spline. This extends prior mouse quadratic and human angle tests in the same release. It is not an independent biological replication. Selected slices, uncertain donor labels and geometry, model form, and intervals conditional on fitted predictions limit the result. The package contains code, source-row extracts, predictions, provenance, checks, and a figure.

Allen Institute · human cortex · spatial connectivity · donor validation · null result