One fitted value per cell reduced log ratio test error by 21% and 11% in two mouse cortical maps of detected spikelets

We tested whether shared cell properties predict unequal fast voltage responses in the two directions of a cell pair. We used 1,133 and 512 pairs from optical voltage maps of layer 1 barrel cortex in two awake male mice. Both cells passed source quality rules, their soma distance was 60 to 400 micrometres, and both directions had detected positive spikelets. A model used one fitted log value per cell. Five outer folds excluded complete cell pairs. Four inner folds selected a ridge penalty. The reference predicted equal normalized responses. Mean squared log ratio error fell by 21.02% and 10.57%. Conditional 95% cell resampling ranges at fixed predictions were 13.23% to 29.03% and -0.94% to 20.32%. These ranges do not estimate variation across mice. A second pair split gave decreases of 20.21% and 12.41%. Keeping all positive estimates without detection flags gave smaller decreases of 4.70% and 0.99%. Full graph fits left 46.38% and 35.56% of squared log ratio size in cycle residuals. These residuals are invariant to fixed scalar source and readout gains. The graph method is established; this work adds a test on dense cellular maps and a compact checked dataset. The result concerns new pairs among the same cells, conditional on full-record selection. Shared noise, normalization, source correction and fast voltage filtering can produce cell patterns and cycle residuals. The fit does not identify physical junction conductance or rectification. The package preserves source extracts, metadata, code, predictions, all random draws, figures and repeat instructions.

electrical coupling · spikelets · functional connectivity · cell properties · direction ratios · graph cycles · voltage imaging · NDNF interneurons · pair prediction · reproducible research