Predicting interaction-specific protein-protein interaction perturbations by missense variants with MutPred-PPI.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Protein-protein interactions (PPIs) are crucial for cellular function, and their disruption by genetic variants can cause disease.
• Existing computational tools often fail to predict interaction-specific (edgetic) effects of variants, especially when stability is unaffected.
• There is a need for scalable computational methods to assess the impact of variants on PPIs.

Purpose of the Study:

• To develop and validate MutPred-PPI, a novel graph attention network for predicting edgetic effects of missense variants.
• To evaluate MutPred-PPI's generalizability and performance against existing methods on diverse datasets.
• To demonstrate the biomedical relevance of MutPred-PPI by analyzing variants from clinical and population databases.

Main Methods:

• Utilized a graph attention network architecture operating on AlphaFold3-based protein complex contact graphs.
• Integrated protein language model embeddings into the graph nodes.
• Performed rigorous evaluation using group cross-validation and benchmark datasets from the IGVF Consortium.

Main Results:

• MutPred-PPI achieved superior performance, with AUCs of 0.85 on seen proteins and 0.72 on unseen proteins in cross-validation.
• The model demonstrated strong generalizability, outperforming all baseline methods.
• Analysis of clinical variants revealed distinct PPI perturbation mechanisms across different disease types, with MutPred-PPI capturing functionally relevant effects.

Conclusions:

• MutPred-PPI is a powerful tool for predicting interaction-specific variant effects, advancing the understanding of molecular mechanisms underlying genetic diseases.
• The study highlights distinct PPI disruption patterns in various diseases, from cancer to neurodevelopmental disorders.
• MutPred-PPI's ability to generalize to unseen proteins underscores its potential for broad application in variant interpretation and precision medicine.