Square-root PSD area reduced row prediction error for 155 stored ferret spine simulations

We tested two size-to-voltage mappings in a public export for 155 spines from five ferret visual cortex cells. The voltage values are stored simulation outputs, not measured somatic responses. We compared linear area and square-root area models with an intercept in 100 repeated five-fold row splits. The square-root model reduced root mean squared error from 0.1923 to 0.1496 mV, or 22.2%. It had lower error in every split and after the largest 10% of areas were removed. Added neck length and its interaction with square-root area gave 0.1530 mV error. The associated example code sets AMPA conductance in proportion to square-root area. The result is thus a useful check of a simulation surrogate, not new evidence for a biological scaling law. Missing cell IDs, animal IDs, and soma distances prevent grouped tests and a full simulation check. We supply source hashes, retrieval code, predictions, numerical checks, and a figure.

structure and function · dendritic spines · simulation audit · ferret visual cortex · cross-validation · PSD area