An artificial neural network model to predict structure-based protein–protein free energy of binding from Rosetta-calculated properties

Abstract

The prediction of the free energy (ΔG) of binding for protein–protein complexes is of general scientific interest as it has a variety of applications in the fields of molecular and chemical biology, materials science, and biotechnology. Despite its centrality in understanding protein association phenomena and protein engineering, the ΔG of binding is a daunting quantity to obtain theoretically. In this work, we devise a novel Artificial Neural Network (ANN) model to predict the ΔG of binding for a given three-dimensional structure of a protein–protein complex with Rosetta-calculated properties. Our model was tested using two data sets, and it presented a root-mean-square error ranging from 1.67 kcal mol⁻¹ to 2.45 kcal mol⁻¹, showing a better performance compared to the available state-of-the-art tools. Validation of the model for a variety of protein–protein complexes is showcased.