With one input repeat, a depth prior gave no clear response forecast gain in 178 cells from one mouse retina

We tested whether dendrite depth improves a light response forecast when one repeat is available. The sample had 178 unique ganglion cells in five fields from one Eyewire II mouse retina. We processed each raw fluorescence repeat separately. The first repeat was the input. The mean of four later normalized repeats was the target. Each test excluded a complete field from fitting and parameter choice. The reference used temporal smoothing and a common response prior. A ridge prior from 20 depth features reduced equal-field mean squared error from 0.353760 to 0.352674, a 0.307% decrease. Four fields improved and one worsened. The central 95% range from resampling five fixed field scores was -1.584% to 1.861%. This range is not a confidence interval for mice. Checks with all input repeats and source processing also had ranges across zero. Four of 199 complete anatomy-label controls reached the recorded gain. A later check found that the common prior reduced error by 3.42% beyond smoothing and shrinkage toward zero. A second numerical implementation reproduced the primary field errors within 2.1e-9. Test forecasts did not depend on later responses from the test field. These results limit the added value of this depth prior under this response score. They do not test all structure or identify a synaptic cause. Code, aggregate results, figures, source checks, and repeat instructions are preserved. Original source arrays remain external under their own terms.

retina · structure and function · dendrite depth · calcium imaging · response estimation · Eyewire II · field exclusion