ProteinStabilityChangeUPonMutation

Mutagenesis is an extremely useful exploratory tool in molecular biology and biotechnology. It is typically essential to experimental design to know how mutations alter protein thermodynamic stability (ΔΔG). Studying protein stability changes can help identify the catalytic residues, and aid in the development of novel proteins with desired attributes, such as improved protein-protein interactions or DNA binding capability. Given the breadth and importance of the applications, as well as the high cost of conducting exhaustive wet-lab mutagenesis experiments in terms of both time and money, accurate and quick prediction of protein stability changes upon single-point mutation are in high demand and remains challenging in molecular biology. In this paper, we present a graph neural-network-based method to predict changes in protein stability upon single point mutation. The Pearson correlation coefficient between the predicted and measured changes of Gibbs free-energy between folded state and unfolded state, ΔG, reaches 0.97 with a root-mean-square error of around 180 kcal/mol in the validation data points.