Across 21 human donors, a ridge test gave no clear added prediction of cortical connections from 15 electrical measures

We tested whether continuous electrical properties help predict detected local connections in human temporal cortex. We used the Planert et al. version 1 data and a strict cohort of 745 neurons, 3,400 directed pair tests and 511 detected connections from 21 donors. Each donor was excluded from regression fitting and penalty selection. The baseline used cell type pairs, soma geometry, access resistance and experimenter labels. Adding 15 electrical measures at each endpoint and their absolute differences reduced mean donor log loss by 0.24%. A central 95% range from resampling fixed donor scores was -1.03% to +1.50%; 10 donors improved. These ranges do not refit the models and are not population confidence intervals. Geometry and acquisition gave a 3.22% gain against a constant probability. Tests with recovered cells, known intact axons, source or target measures alone, extra quadratic terms, and no type labels gave uncertain gains. Ten alternate donor splits gave -0.07% to +1.35%; one fixed-score range was positive. A later indicator-scaling check gave -0.09%. The main test gave no clear added prediction gain. It does not show that electrical properties and connectivity are independent. Surgical tissue, response detection, one region and layer group, joint source cell labels, and the fixed model limit the result. The package gives new code, source receipts, aggregate results, model states, figures, verification and repeat instructions. Original observations remain external.

human cortex · connectomics · synaptic connectivity · intrinsic physiology · donor exclusion · ridge regression · prediction · negative result