Data-extracted gating kinetics yield parameter-efficient, predictive models of four cardiac ion channels

Cardiac ion-channel models underpin in silico drug-safety assessment, yet the standard route, global fitting of Markov schemes to voltage-clamp data, yields non-identifiable parameters. Here we invert the workflow: for four cardiac channels (hERG/I_Kr, Na_v1.5/I_Na, Ca_v1.2/I_CaL, KCNQ1+KCNE1/I_Ks) we extract gating time constants, steady-state availability and delays from public voltage-clamp data as voltage-dependent tables, test each for constancy, and assemble models with one free per-cell conductance and, where data demand it, one midpoint shift. Time-constant tables prove portable across cells (CV 0.13--0.19, hERG deactivation; 0.029, Na_v1.5 inactivation at threshold); voltage midpoints do not. The pattern repeats in all four channels and in an independent five-temperature programme (670 wells). Without global fitting, the models forward-predict held-out protocols, including action-potential clamp and 65 of 68 CiPA drug-screening traces. A Fisher-information analysis of a 13-parameter Markov model shows the data identify only modal coordinates (six null directions; condition number 7.4 x 10^12). Across the tested drug panels, in all four channels drugs reject pure pore block and modulate steady-state occupancy and inactivation tables, not kinetics. A fifth channel (K_v4.3) reproduces all published anchors; a candidate condition-dependent inactivation-timescale shift is registered. Every governing quantity is either measured directly or an explicitly registered assumption; cell-to-cell variability is localised, not averaged away.