Genetic algorithm-based personalized models of human cardiac action potential

Abstract

We present a novel genetic algorithm-based solution to determine the set of cardiomyocyte model parameters based on experimental human action potential (AP) recordings. The novel approach is based on AP waveform dependence on the heart rate. In order to find the steady-state solution, optimized parameters include conductivities of ionic channels and exchangers augmented by slow variables, intracellular sodium concentration and sarcoplasmic reticulum calcium load. The algorithm is enhanced by a novel mutation operator, based on Cauchy distribution along a random direction in parameter space. We also demonstrate that increasing the number of elite organisms up to 7% results in faster convergence. Test runs indicate that algorithm error is below 5% for IKr, 7% for IK1 and INa and 13% for ICaL. Experimental signal-to-noise ratio above 28 dB was sufficient for high quality algorithm performance. The algorithm validation using optical mapping recordings of human ventricular AP demonstrated low error that was below 6 mV for AP waveform and less than 16 ms for AP duration. Further validation of the personalized models was done using mRNA expression profile of two donor hearts. The mRNA-based model reproduced AP waveform dependence on cycle length with 13 mV accuracy and AP duration with 20 ms accuracy.