In three vangl2 zebrafish maps, a contact area rule from seven normal maps cut synapse log loss by 41% versus a constant rule

We tested whether a contact rule fitted to normal zebrafish can predict ribbon connections in vangl2 mutants. The main sample had 339 positive afferent contacts from 91 hair cells in seven normal maps and three mutant maps. A logistic rule used contact area alone. Normal tests excluded each complete map in turn. Mutant tests used only normal maps for fitting. No age, direction, position, test synapse label or test ribbon total entered the main rules. Mean mutant log loss was 0.393426, versus 0.665821 for a constant rule fitted to normal maps, a 40.9% decrease. Mean mutant AUC was 0.916954. Area reduced loss in each mutant map. However, it underestimated connection prevalence by a mean of 9.17 percentage points. Adding relative area reduced mean normal loss by 0.015958, with sign-flip reference p = 0.375, and helped in only two of three mutants. The first analysis used source ages. We removed age after finding that some source age labels used ribbon volume. Both stages are preserved. A secondary equal-gain cosine model gave larger relative direction-vector errors in mutants despite similar edge loss. This is a conditional model, not measured nerve activity. The source already established selective wiring. We add a small transfer benchmark, its calibration limit, code, source data and provenance, every forecast, figures and repeat instructions. Three mutant maps do not support a population claim.

connectome · zebrafish · neuromast · contact area · synapse prediction · vangl2 · transfer prediction · sensory direction · calibration