Reliable prediction of protein-protein binding affinity changes upon mutations with Pythia-PPI

Protein-protein interactions are essential for numerous biological functions, and predicting binding affinity changes caused by mutations is crucial for understanding the impact of genetic variations and advancing protein engineering. Although machine learning-based methods show promise in improving the prediction accuracy, the scarcity of experimental data remains a significant bottleneck. Here, we utilized multi-task learning and self-distillation to overcome the data limitation and improved the accuracy of deep learning-based protein binding affinity prediction. By incorporating a mutation stability prediction task, the model achieved state-of-the-art accuracy on the SKEMPI dataset and was subsequently used to predict binding affinity changes for millions of mutations, generating an expanded dataset for self-distillation. Compared with prevalent methods, Pythia-PPI increased the Pearson correlation between predictions and experimental data from 0.6447 to 0.7850 on the SKEMPI dataset, and from 0.3654 to 0.6051 on the virial-receptor dataset. These findings demonstrated Pythia-PPI to be a valuable tool for analyzing the fitness landscape of protein-protein interactions. We provided a web server at https://pythiappi.wulab.xyz for easy access to Pythia-PPI.