MPA-Pred: A machine learning approach for predicting the binding affinity of membrane protein-protein complexes

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Membrane protein-protein interactions are crucial for cellular functions like signaling and transport.
• Binding affinity dictates these interactions, but specific prediction methods for membrane proteins are lacking.
• Existing methods for general protein-protein interactions do not adequately address membrane protein complexities.

Purpose of the Study:

• To develop a specialized method for predicting binding affinity in membrane protein-protein complexes.
• To identify key structural and sequence-based features influencing membrane protein binding affinity.
• To create a user-friendly tool for large-scale prediction and drug design support.

Main Methods:

• Collected experimental binding affinity data for 114 membrane protein-protein complexes.
• Derived and analyzed structure- and sequence-based features, focusing on interface residues and interactions.
• Developed a machine learning model, MPA-Pred, for binding affinity prediction.

Main Results:

• Identified aromatic/charged residues and aromatic-aromatic/electrostatic interactions as critical for binding affinity.
• MPA-Pred achieved a high prediction accuracy with a correlation of 0.83 and MAE of 0.91 kcal/mol on 114 complexes.
• The method's performance was validated using a jack-knife test.

Conclusions:

• MPA-Pred accurately predicts membrane protein-protein complex binding affinity.
• The findings highlight the importance of specific residue types and interactions at the interface.
• The developed web server facilitates large-scale predictions, potentially accelerating drug discovery efforts.