Biophysical Society Thematic Meeting | Canterbury 2023

Towards a More Perfect Union: Multi-Scale Models of Muscle and Their Experimental Validation

Poster Abstracts

2-POS Board 2 USING NEURAL NETWORKS TO PREDICT PATHOGENICITY OF VARIANTS IN THE CARDIAC THIN FILAMENT Ananya Chakraborti 1 ; Allison B Mason 1 ; Jil C Tardiff 2 ; Steven D Schwartz 1 ; 1 University of Arizona, Department of Chemistry and Biochemistry, Tucson, AZ, USA 2 University of Arizona, Department of Biomedical Engineering, Tucson, AZ, USA Single point mutations to the different proteins of the cardiac thin filament complex can be associated with loss of functionality, generally resulting in genetic cardiomyopathies. Currently, linking genotype to phenotype to determine the pathogenicity of the variants is difficult. Previously, our group used the molecular dynamics simulations to determine the baseline of pathogenic variations found via analysis of computational observables. The pathogenic criterion was then used to predict the pathogenicity of variants of unknown significance (VUS). In this study, we aimed to create a convolutional neural network (CNN) model to predict the pathogenicity of VUS using the results of the molecular dynamics simulations of the full computational model of cardiac thin filament. The basic algorithm consists of several 2D convolution and max pooling layers to reduce the dimensionality of the molecular dynamics data, followed by a dense network and rectified linear unit (ReLU) activation function, finally giving the output. Our neural network model will be able to predict whether a new set of VUS on cardiac troponin T, cardiac troponin I and tropomyosin are benign or pathogenic, with the pathogenic predictions further split into either hypertrophic or dilated cardiomyopathies. These predictions will aid the clinicians to prescribe the appropriate clinical treatment.

79