Source-peer delay calibration gave no clear gain over disjoint peers in three screened Allen cortical slice groups

We tested whether measured delays at other outputs of a neuron improve prediction of one unmeasured output. We used Allen Synaptic Physiology r2.1 and fit a cell-label and distance model with five folds that held out inferred donors. We then compared source-peer residual calibration with calibration from connections in the same recording that used neither cell. The screened sample kept onset delays from 0.2 to 5 ms. Mouse excitatory, mouse inhibitory, and human excitatory samples had 71, 193, and 39 target connections. All three conditional 98.75% donor bootstrap intervals for the added gain included zero. Mouse inhibitory gain was 0.01830 squared log units, with interval -0.00315 to 0.04336. Source calibration did improve on the uncalibrated model in that group. Without the delay screen, the source-versus-disjoint interval was above zero, so the conclusion depends on selection. Human inhibitory data had only two inferred donors and remain descriptive. The package supplies a sparse-measurement benchmark, peer IDs, predictions, source hashes, checks, and a figure. Small selected circuits, inferred donors, fixed calibration strengths, and uncertainty conditional on fitted models limit the result.

synaptic delay · sparse measurement · circuit structure · Allen Institute · calibration