With independent site events and unit effects, decoders that ignored sharing lost 9.76–10.50% of optimal squared SNR at the median in two adult C. elegans input models

We tested how shared synapse site events affect a linear readout of direct neural inputs. Two adult C. elegans brain maps supplied 9,591 and 9,180 source contrasts. Source changes had equal total mean target input. Each site had one independent event and unit positive effects on its listed targets. A decoder that used separate contact events lost 9.76% and 10.50% of optimal squared signal-to-noise ratio at the median. Both event models gave equal target marginal laws. Common source event correlation of 0.1 reduced median loss to 3.70% and 3.84%. Geometric target weights reduced the loss to 4.71% in one adult map. A correlation template from the other adult improved modeled separation in 86.67% and 89.03% of contrasts when test means and target variances were known. Exact event enumeration, raw pair counts, fresh source hashes and a complete repeat checked the results. These are conditional input model scores, not measured neural information or classification accuracy. The package preserves all scores, compact facts, source provenance, figures, proofs, code and repeat instructions.

connectome · polyadic synapse · shared release · linear decoder · signal-to-noise ratio · Caenorhabditis elegans · source noise · anatomical model · reproducible benchmark