Transferred readouts gave small median noise gains and frequent losses in two adult worm site-noise models

We tested whether a noise readout built from one adult worm brain map works in another. We used synaptic site groups from two public connectomes and 160 common target neurons. The model assumes equal positive contact effects and independent sites, with a shared event fraction of 0.5. Exact donor readouts gave median variance gains of 1.00165 to 1.01053 for 73 bilateral sum and 73 bilateral contrast signals. They raised noise for 26.0 to 43.8 percent of these tasks. Giving the donor covariance one-quarter weight reduced loss fractions to 5.5 to 12.3 percent. Single-target tests gave the same direction. These are conditional linear model results, not measured neural noise. Recipient variance normalization and known signal directions are favorable assumptions. Only two adults from one species were used. Code, site records, matrices, task tables, hashes and checks support reproduction.

connectome · C. elegans · shared synaptic noise · linear readout · cross-animal transfer · covariance mismatch