With eight group means available, shared partner models did not reduce trace test error in 84 zebrafish hindbrain neurons from one fish

We tested whether exact synaptic partners add information about individual calcium response shapes after known function groups. We used all released input and output tables for 301 reconstructed units from one larval zebrafish. The test included 84 imaged hindbrain neurons and four visual stimuli. Four function groups, each split by hemisphere, gave eight groups. A fixed narrow penalty grid gave a 0.1104% error decrease after input and output partner profiles were added. However, group means alone had lower error, and every main fit chose the largest allowed penalty. A later check added larger penalties and a group mean option. The input, output, and combined partner models selected group means alone in all 84 excluded-neuron tests, with mean squared error 0.346160. Including synapses below the source size cut-off gave error 0.346207, or 0.0135% above the group mean error. Only 191 of 3,486 cell pairs shared a retained input partner. The result limits added prediction in this sparse graph and conditional model. It does not test unknown function labels or establish a lack of synaptic function. The package preserves source tables, sparse graphs, trial data, code, plans, predictions, figures, and provenance.

zebrafish · connectome · structure and function · calcium imaging · shared input · kernel regression · regularization · evidence accumulation · reproducible research