PLM-interact: extending protein language models to predict protein-protein interactions

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in biology

Background:

• Protein structure prediction from sequence is accurate, but protein-protein interaction (PPI) prediction remains challenging.
• Existing methods using protein language models (PLMs) often overlook the physical interaction context.
• There is a need for advanced computational methods to accurately predict complex biomolecular relationships.

Purpose of the Study:

• To evaluate and adapt protein language models (PLMs) for predicting protein-protein interactions.
• To develop a novel approach, PLM-interact, that jointly encodes interacting protein pairs.
• To assess the model's performance on cross-species interaction prediction, mutation effects, and virus-host interactions.

Main Methods:

• Retraining protein language models (PLMs) specifically for protein-protein interaction prediction.
• Developing PLM-interact, a method that jointly encodes protein pairs, inspired by natural language processing's next-sentence prediction.
• Fine-tuning PLM-interact to predict the impact of mutations on protein interactions.
• Evaluating performance on a cross-species benchmark dataset and virus-host interaction prediction.

Main Results:

• PLM-interact achieves state-of-the-art performance on a cross-species protein-protein interaction prediction benchmark.
• The model demonstrates high accuracy when trained on human data and tested on diverse species like mouse, fly, worm, E. coli, and yeast.
• A fine-tuning method successfully detects mutation effects on protein interactions.
• The approach outperforms existing methods in predicting virus-host protein interactions.

Conclusions:

• Protein language models can be effectively extended to learn complex biomolecular relationships, including protein-protein interactions, directly from sequences.
• PLM-interact represents a significant advancement in predicting protein-protein interactions and related biological phenomena.
• This work highlights the potential of large language models in deciphering intricate molecular interactions in biology.