Target terms cut decay test error by 12% and 7% in two mouse maps, with cell refit ranges that included zero

We tested whether other connections of a cell help predict inhibitory response decay in two optical maps of mouse barrel cortex layer 1. The source paper had reported a target association with decay. We extended that result with NPY label, distance, and source drive controls, training-only penalty selection, and tests that excluded complete cell pairs. The screen retained 639 and 440 independently fitted responses. Target terms reduced squared log test error by 11.88% and 6.93%. Source terms reduced it by 1.99% and 0.28%; after target terms entered, the source gains were 0.45% and 0.83%. Target point gains stayed positive across planned precision, distance, amplitude, and split checks. With predictions fixed, central 95% cell resampling ranges for target gain were 1.94% to 22.03% and -2.00% to 14.96%. Refit ranges were -11.61% to 21.81% and -15.54% to 19.19%. These conditional ranges are not confidence intervals across mice. The maps measure optical responses, not verified anatomical contacts. Shared waveform corrections can cause cell patterns. The result supports a modest target prediction pattern in these maps, with uncertain gain under cell refits. The package preserves source tables, provenance, code, predictions, model states, resampling draws, figures, and offline repeat instructions.

functional connectomics · inhibitory postsynaptic potential · decay · layer 1 · NPY · cell identity · prediction · uncertainty