ThermoNet

ThermoNet is a software tool that uses deep learning algorithms to predict changes in protein thermodynamic stability (ΔΔG) due to point mutations. It does so by treating protein structures as multi-channel 3D images and leveraging the image processing power of convolutional neural networks (CNNs). The input to ThermoNet is constructed as multi-channel voxel grids based on biophysical properties derived from raw atom coordinates. The tool was trained using a curated dataset that accounts for protein homology and balances direct and reverse mutations, providing a framework for addressing biases in previous methods. ThermoNet performs comparably well to existing methods on the widely used Ssym test set and accurately predicts the effects of both stabilizing and destabilizing mutations. Its practical utility was demonstrated by predicting ΔΔGs for two clinically relevant proteins, p53 and myoglobin, and for pathogenic and benign missense variants from ClinVar.