With a fixed optical kernel, nearby cell wiring scores raised response magnitude test error by 0.16% in 93 C. elegans recordings

Cells near an intended optical stimulation target could send signals to distant readout cells. We tested whether their reference wiring improves response prediction. We used 9,774 animal-direction means from 590 events in 93 immobilized C. elegans recordings. Readouts were at least 10 planar micrometres from the intended source. Candidate cells received Gaussian weights with planar full width at half maximum of 3.1 micrometres and axial weights of one. Two scores summed their outgoing chemical and gap counts to each readout. The base model already had intended-source wiring, exact neuron labels, image geometry, recording controls, candidate count, weight mass and unweighted candidate wiring. Nested ridge validation excluded complete animals. Adding the nearby wiring scores raised mean squared response magnitude error from 0.186764 to 0.187062, an increase of 0.160%. The conditional 95% range for error decrease was -0.292% to -0.043%; it held predictions fixed and omitted training uncertainty. A second animal partition gave a 0.032% error increase with a range that included zero. No declared check gave a conditional range wholly above zero gain. The result limits these scores under the assumed kernel; it does not exclude off-target activation. The package preserves new geometric scores, source hashes, original-data rebuild code, model code, all predictions, control results and figures.

Caenorhabditis elegans · connectome · signal propagation · optogenetic stimulation · off-target activation · recording geometry · animal validation · calcium imaging