Synaptic delay gave no clear variability prediction gain in four Allen slice groups

We tested whether synaptic delay improves predictions of resting response variability after cell labels, cell-body distance, response size, noise, and event count are known. We used 1,515 connections from the Allen Synaptic Physiology r2.1 small database. Models held out whole inferred donors. A log delay term gave no clear gain in mouse excitatory, mouse inhibitory, human excitatory, or human inhibitory groups. Mouse gains were -0.18% and 0.05% of reference mean squared error. Human inhibitory gain was 21.63%, but its sample had only 20 connections from 11 inferred donors and its adjusted conditional interval included zero. Lower regularization, removal of the response-size control, and removal of extreme outcomes did not give a clear positive gain. The result limits the use of a simple delay term as a reliability predictor in these selected slices. It does not show independence or identify a release mechanism. Missing measurements, fit constraints, inferred donor labels, and conditional intervals limit inference.

Allen Institute · synaptic delay · response variability · donor validation · null result