Mirror response data cut mean test error by 0.43% in 103 C. elegans worms kept out of fitting

We tested whether a measured mirror response helps predict a response in a different C. elegans worm. The panel had 24,328 animal-direction means from 103 worms. Five outer groups kept whole worms out of fitting and tuning. A reference used exact cell identities, own and mirror wiring, and exact-direction response calibration from other worms. Combining exact and mirror residual means cut equal-animal mean squared error from 0.531640 to 0.529337, a 0.433% reduction. A central 95% whole-worm bootstrap range was 0.240% to 0.627%. It held predictions fixed. Reverse-direction data gave a 0.400% reduction. The mirror advantage over reverse data had a range across zero. A denser-grid audit found that mirror values added information beyond mirror counts. The common pre-stimulus panel gave no clear gain. When the exact direction had no training measurement but its mirror did, the gain was -0.041%, with a range from -1.040% to +0.864%. Thus the overall gain does not establish a method to fill unmeasured directions. All seven specified post-response checks kept positive mean gains. The other animal split had a range very close to zero. These are conditional results for stated estimators and calcium windows. They do not identify a synaptic mechanism or equal neural function. The package preserves code, source versions and hashes, compact data, all predictions, model states, ranges, figures, and repeat instructions.

connectome · structure and function · Caenorhabditis elegans · bilateral symmetry · signal propagation · animal validation · response prediction · partial pooling