Trained peer weights gave no mean 50 Hz prediction gain in selected Allen slice circuits

We tested completion of a missing 50 Hz response measure from connections that share a source or target cell. Exact cell-label matching in Allen Synaptic Physiology r2.1 left 78 mouse connections from 35 inferred animals and 12 human connections from six inferred donors. We learned a bounded peer weight on other donors. Mouse baseline, source, and target mean squared errors were 0.05831, 0.05852, and 0.05893. Human errors were 0.07145, 0.09426, and 0.09734. Neither trained method gave a mean gain. Most mouse weights were zero. This tests one simple method in sparse, selected slice circuits. It does not establish equal source and target roles or a lack of wiring information. Few human donors, fit error, and cell-type mixtures limit transfer. Code, aggregate evidence, a figure, hashes, and retrieval instructions support reproduction. External record-level data are not bundled.

Allen Institute · short-term plasticity · peer prediction · shrinkage · circuit structure · null result