Predicting pathogenicity of variants in the cardiac thin filament using neural networks.

Abstract

Single amino acid variations to the various proteins of the cardiac thin filament have been linked to a loss of functionality and have shown to result in genetic cardiomyopathies. Currently, there are significant difficulties in linking genotype to phenotype to determine pathogenicity of variants, limiting the ability of clinicians to intervene. Previously, we used molecular dynamics simulations of pathogenic variants on cardiac troponin T and tropomyosin to determine a baseline of pathogenic changes induced in computational observables. The pathogenic baseline was then used to predict the pathogenicity of several variants of unknown significance. In the current study, a convolutional neural network (CNN) is being created to predict the pathogenicity of variants of unknown significance using results of molecular dynamics simulations of the all-atom computational model of the full human cardiac thin filament. The neural network will be able to predict whether a new set of variants of unknown significance on cardiac troponin T, cardiac troponin I, or tropomyosin are benign or pathogenic and further split the pathogenic predictions into either hypertrophic or dilated cardiomyopathy inducing. The CNN will consist of several convolution and pooling layers to reduce the dimensionality of the molecular dynamics data followed by a fully dense network.