Separate gains gave 69% more squared log repeat error in 18 human cortical connections at low calcium

We tested how to assign a membrane-state response gain to a known connection. We used two released repeats for 18 human cortical connections at 1.2 mM external calcium. Each gain was the Depol response mean divided by the Control response mean. We compared an individual gain with a common gain from other record prefix groups. We predicted each repeat from the other repeat and measured squared error on the natural log scale. The individual rule gave error 0.11598, versus 0.06852 for the common rule, a 69% increase. Five record prefix groups limited the uncertainty test. With predictions fixed, the 95% group resampling interval for the error ratio was 1.24–2.93. After full refits, it was 0.79–3.68. Thus, transfer of the point difference to new groups was uncertain. In ten other connections before drug exposure, the individual rule gave 6% less error, with a wide interval. The package preserves two unchanged source tables, predictions, source hashes, code, and figures. This is a calibration test on selected slice recordings. It does not show that true gains are equal or unrelated to anatomy.

human cortex · synaptic transmission · membrane potential · connectome calibration · repeat prediction · measurement error · group resampling · open data