Accurate protein-protein interactions modeling through physics-informed geometric invariant learning.

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Accurate prediction of protein-protein interactions is crucial for understanding biological functions.
• Existing methods like AlphaFold face limitations in predicting complexes, engineered proteins, and antigen-antibody interactions due to sparse co-evolutionary data.

Purpose of the Study:

• To develop ProTact, a novel SE(3)-invariant geometric graph neural network for enhanced protein-protein contact prediction.
• To improve the accuracy of protein docking pose approximation.

Main Methods:

• ProTact utilizes physics-informed geometric complementarity and trigonometric constraints as inductive biases.
• It employs a modulated key point matching algorithm for docking pose approximation.
• The method is applicable to both experimental and predicted monomer structures.

Main Results:

• ProTact significantly outperforms state-of-the-art methods on benchmark datasets, showing >30% improvement in Precision@10 for CASP and DIPS-Plus.
• It maintains a competitive edge on challenging unbound complexes.
• When used with AlphaFold3, ProTact improves default confidence scores by over 30% in low-MSA contexts.

Conclusions:

• ProTact offers a robust framework for predicting protein-protein interactions, overcoming limitations of current methods.
• The approach enhances understanding of protein interactions, functions, and facilitates protein design.
• This work advances computational structural biology and drug discovery efforts.