Reverse excitation reduced ridge prediction error for inhibition by 36.9 percent in 66 reciprocal mouse V1 pairs with visual controls

We tested whether reverse synaptic strength adds information beyond measured visual response similarity. We used 66 reciprocal PV and pyramidal cell pairs from 27 mice in a public V1 dataset. We recalculated excitatory and inhibitory postsynaptic potentials from accepted slice trials and matched each pair to its in vivo visual responses. A baseline model used visual similarity, distance, response variance measures, and postsynaptic voltages. Each test mouse was absent from model fitting and penalty selection. Adding the reverse excitatory potential reduced mean mouse squared log inhibition error from 0.11819 to 0.07464, or 36.9 percent. Error fell in 21 of 27 mice. The 95 percent mouse resample range was 12.0 to 55.3 percent with predictions fixed. Among 37 pairs with several partners per PV cell, the adjusted within-cell correlation was 0.557. A slope refit after mouse resampling gave a range of 0.201 to 1.386. The stricter visual response sample had 31 pairs and an uncertain 15.3 percent gain. A later optical separation check gave a 30.1 percent gain in 60 pairs. These results extend known reciprocal weight matching with a test on excluded mice and joint visual controls. They do not identify a plasticity mechanism or remove error in visual similarity. The package preserves new numerical tables, code, source hashes, model states, figures, and repeat instructions.

visual cortex · parvalbumin · reciprocal connections · synaptic strength · visual response similarity · grouped prediction