Depth overlap gave a 1.27% mean test error decrease for residual calcium correlation in 147 ganglion cells from one mouse retina

We tested whether overlap of dendrite depth profiles adds prediction of shared trial variation in retinal cells. We used matched anatomy and five repeated light responses from Eyewire II. The main sample had 147 ganglion cells and 2,206 pairs in five fields from one mouse retina. We aligned each trial to its recorded stimulus trigger. Each test excluded one field from model fitting. The baseline used soma distance, mean response similarity, and response quality. Adding depth overlap reduced mean squared test error by 1.27%. Four fields improved and one worsened. The central 95% range from resampling five fixed field scores was -0.73% to +2.83%. It is not a confidence interval for mice. None of 199 anatomy-label controls reached the recorded gain. The gain fell to 0.14% after conditioning each pair on the activity of the other cells in its field. It was 0.58% after removing trial means and 0.59% after removing a linear trial trend. The result limits the added value of this depth feature under the tested conditions. It does not identify a synaptic noise source. Code, aggregate results, figures, source hashes, and reproduction instructions are preserved. Source arrays remain external under their original terms.

retina · structure and function · dendrite depth · calcium imaging · trial residuals · shared variation · Eyewire II · reproducibility