EuDockScore: Euclidean graph neural networks for scoring protein-protein interfaces

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in protein science

Background:

• Protein-protein interactions are crucial for biological processes, but predicting their structures computationally remains challenging.
• Experimental structure determination is resource-intensive, necessitating accurate computational methods.
• Scoring functions are vital for evaluating protein complex models generated by docking and deep learning.

Purpose of the Study:

• To develop novel, high-performance scoring functions for protein-protein interactions.
• To create specialized models for antibody-antigen complex assessment and reranking.
• To leverage advanced Euclidean graph neural network architectures for improved accuracy.

Main Methods:

• Utilized cutting-edge Euclidean graph neural network architectures.
• Developed EuDockScore for general protein-protein interactions.
• Created EuDockScore-Ab for antibody-antigen docking and EuDockScore-AFM for reranking AlphaFold-Multimer outputs.

Main Results:

• Presented improved scoring functions (EuDockScore, EuDockScore-Ab) for assessing protein-protein interfaces.
• Demonstrated the utility of EuDockScore-AFM in reranking large sets of antibody-antigen complex predictions from AlphaFold-Multimer.
• Achieved enhanced accuracy in evaluating protein complex candidate structures.

Conclusions:

• The developed EuDockScore models offer significant improvements in scoring protein-protein interactions.
• EuDockScore-AFM provides an effective solution for filtering and prioritizing antibody-antigen complex predictions.
• These advancements facilitate more accurate and efficient computational modeling of protein complexes.