Chemical arrow reversal left profile similarity unchanged with equal input and output weights in five C. elegans maps

We tested whether one wiring profile model retains the direction of chemical connections. Its neuron similarity is the mean of output, input and electrical similarity matrices. Full chemical arrow reversal exchanges the output and input terms. We prove that their equal weights leave the pooled similarity fixed. We checked five C. elegans maps and replayed 21 published model cases based on a 6,270-direction calcium response panel. The largest prediction change was 2.6e-15 when only the added wiring profile graph was reversed. Local wiring controls stayed fixed. We then tested two separate 256-feature models with both endpoint groups excluded from response training. One used source output profiles and target input profiles. The other exchanged these roles. In the primary split, the first model had 0.84% more error. Its descriptive endpoint weight range for error reduction was -2.53% to 0.17%. It had lower error in 0 of five planned splits. The pooled feature gain therefore cannot identify the chemical orientation of the graph. This is a limit of the tested representation. It does not show that biological signalling ignores chemical direction. The tests use pooled calcium means from one atlas and fixed ridge model families. Code, input arrays, source records, model states and figures are included.

connectome · Caenorhabditis elegans · signal propagation · graph kernels · chemical direction · reproducibility