Reverse synaptic strength gives no clear prediction gain across mouse donors in Allen reciprocal-pair data

We tested whether the strength of one direction of a reciprocal connection predicts the other direction in new animals or donors. We used resting-state postsynaptic potential amplitudes in the Allen Synaptic Physiology r2.1 small database. The reference model used cell labels, layer, connection types, and cell-body distance. Validation kept all records from each inferred donor together. In 283 mouse reciprocal pairs from 211 inferred animals, reverse strength did not lower prediction error. Three near-zero fitted amplitudes caused large log errors. A check with both amplitudes at least 10 microvolts kept 277 pairs from 208 animals. Log-amplitude RMSE was 1.0111 for the reference model and 1.0121 with reverse strength. The mean squared error gain was -0.0021, with a conditional donor bootstrap interval of -0.0202 to 0.0144. A simpler reference model gave a small gain with an interval that included zero. Type-specific slopes did not improve the pooled result. Only 19 human reciprocal pairs met the main filters. These tests do not show that reciprocal strengths are independent. They show that a reverse-strength predictor gave no clear added value across donors under the tested selection and models.

reciprocal connections · synaptic strength · Allen Institute · grouped validation · null result