With a fixed training weight of 0.5, six larval maps cut allocation error by 32% for 130 groups in one adult C. elegans map

We tested whether anatomical allocation weights at shared synaptic sites repeat across C. elegans animals. The weights come from diffusion simulated in reconstructed geometry. They are not measured transmission strengths. Six larval maps supplied the training data. One adult map supplied the primary test. For each group of one source and two targets, we predicted the mean pair allocation ratio. A group needed at least three shared sites in the adult and in each of at least two larvae. The fixed rule moved halfway from equal allocation toward the larval mean. With equal total weight for each source class, mean squared error fell from 0.015017 to 0.010207, a reduction of 32.03%. The test had 130 groups in 33 source classes. It covered 855 of 2,480 adult sites with two valid panel targets. The training and adult means had correlation 0.557. A source class resampling range was 11.81% to 47.09% reduction. A wider check across all three neuron roles gave -34.02% to +62.74%. Thus, the gain is a result for these maps and does not establish transfer across adult animals. Fixed count, volume and other-source references gave higher error than equal allocation. All six larval checks gave positive gains against equal allocation, but they shared training animals. Geometry and common annotation rules remain possible causes. The package preserves code, source hashes, derived inputs, all predictions, all checks, figures and independent audits.

connectome · Caenorhabditis elegans · polyadic synapse · allocation weights · anatomical prediction · reproducibility