Learning the language of protein-protein interactions

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Protein Language Models (PLMs) are effective for individual protein properties but struggle with protein-protein interactions (PPIs).
• Understanding PPIs is crucial for cellular processes and disease mechanisms.
• Existing models lack scalability and contextual representation for complex protein interactions.

Purpose of the Study:

• Introduce MINT, a novel PLM designed to model sets of interacting proteins.
• Enhance the understanding of PPIs, their binding affinities, and mutational effects.
• Develop a scalable and contextual model for complex protein interactions.

Main Methods:

• Unsupervised training of MINT on a large, curated PPI dataset from the STRING database.
• Evaluation of MINT's performance on diverse PPI-related tasks, including binding affinity prediction and mutational effect estimation.
• Benchmarking MINT against existing PLMs and specialized models for complex interactions.

Main Results:

• MINT demonstrates superior performance in various PPI tasks compared to existing PLMs.
• MINT excels at modeling interactions in complex protein assemblies, antibody-antigen interactions, and T cell receptor-epitope binding.
• MINT's predictions for mutational impacts on oncogenic PPIs and antibody cross-neutralization align with experimental data.

Conclusions:

• MINT is a powerful tool for modeling complex protein-protein interactions.
• The model has significant implications for biomedical research, therapeutic discovery, and understanding disease mechanisms.
• MINT offers a scalable and contextual approach to PPI analysis.