PPAC: Predicting Protein-Protein Affinity Changes Induced by Amino Acid Mutations Using Protein Large Language Models

Area of Science:

• Computational Biology
• Biophysics
• Drug Discovery

Background:

• Accurately predicting the impact of amino acid mutations on protein-protein binding free energy is crucial for drug design and functional biology.
• Traditional methods often face limitations in capturing complex mutational effects.

Purpose of the Study:

• To develop and validate a novel approach using Protein Large Language Models (PLMs) for predicting mutational effects on protein-protein binding free energy.
• To establish a new state-of-the-art (SOTA) predictive model for protein interaction analysis.

Main Methods:

• Utilized three advanced PLMs (Esm2, EsmC, ProtT5) to generate sequence-based protein representations.
• Integrated PLM representations into seven distinct model architectures and selected the optimal combination via 5-fold cross-validation.
• Trained the final model (PPAC) on a large-scale dataset and evaluated it on 9,558 data points.

Main Results:

• The PLM-based method significantly outperformed traditional approaches in predicting mutational effects.
• Achieved state-of-the-art (SOTA) predictive performance.
• The PPAC model demonstrated high precision and identified key residues critical for protein interactions.

Conclusions:

• PLMs offer a powerful tool for characterizing protein variants and predicting binding free energy changes.
• The developed PPAC model provides a significant advancement in protein interaction modeling and drug design.
• This approach enhances the identification of critical residues involved in protein binding, offering valuable insights for functional biology.