Separate regional calcium filters gave 7.9% more glutamate test error in calibrated UV cones from four zebrafish during repeated light steps

We tested whether retinal region helps to predict glutamate fluorescence from calcium fluorescence in UV cones. We used 56 cones from four zebrafish in Dryad version 3 of the Schröder et al. data. Early repeats set each cone's baseline, linear drift and scale. We scored later repeats of 3 s dark and 3 s bright light steps. We excluded each fish from filter fitting and penalty selection. A common ridge filter used calcium from 1 s before to 1 s after the glutamate time. Separate regional filters gave mean squared error of 0.057915. The common filter gave 0.053663. With equal fish weights, error increased by 7.9%. All four fish had more error. A common light-response template gave less error, at 0.032875. Regional templates gave 12.5% more error. Alternate calibration lengths and calcium windows kept the mean error increase. A blend selected the common filter for every test fish. Shape-only scores were mixed. Rescaling fitted predictions gave a 1.6% gain. Refitting unit waveforms gave 2.1% more error. Separate electron microscopy cells show regional ribbon differences. They are not paired with the functional cells. This benchmark does not test absolute release, new stimuli or the source nonlinear pool model. It cannot assign a causal effect to ribbon geometry. Four fish limit population inference. The package preserves the source files, their hashes and terms, original code, predictions, scores, figures and repeat instructions. Independent parsing and fits checked the primary results.

UV cones · ribbon synapses · structure and function · calcium · glutamate · zebrafish · animal exclusion · prediction benchmark