MutPred-PPI predicts whether a missense mutation disrupts a specific protein–protein interaction (PPI). Given the three-dimensional structure of a protein complex and one or more missense variants, MutPred-PPI outputs a score from 0 (interaction preserved) to 1 (interaction disrupted) for each variant–partner combination.
The method integrates structural context from the protein complex with sequence-level information from protein language model embeddings, processed by an ensemble of graph neural networks trained on curated PPI perturbation data.
Stewart, R. et al. (2026). Predicting interaction-specific protein–protein interaction perturbations by missense variants with MutPred-PPI. In Proceedings of RECOMB 2026. doi:10.64898/2025.12.20.695738
@inproceedings{stewart2026mutpred-ppi,
title = {Predicting interaction-specific protein--protein interaction
perturbations by missense variants with {MutPred-PPI}},
author = {Stewart, Ross and Laval, Florent and Coppin, Georges and
Spirohn-Fitzgerald, Kerstin and Tixhon, Maxime and Hao, Tong and
Calderwood, Michael A and Mort, Matthew and Cooper, David N and
Vidal, Marc and Radivojac, Predrag},
booktitle = {Proceedings of the 30th Annual International Conference on
Research in Computational Molecular Biology (RECOMB)},
year = {2026},
doi = {10.64898/2025.12.20.695738}
}
Training and evaluation data are available on Zenodo: doi:10.5281/zenodo.17645488 .
The standalone MutPred-PPI tool is available at github.com/rosstewart/mutpred-ppi .
MutPred-PPI is released under the MIT License. Users must comply with the terms of any structural data used as input. AlphaFold3 outputs are subject to AlphaFold3 Output Terms of Use (non-commercial use only).