Graphical models of protein-protein interaction specificity from correlated mutations and interaction data

Area of Science:

• Computational Biology
• Biochemistry
• Bioinformatics

Background:

• Protein-protein interactions are crucial for cellular functions.
• Interaction specificity arises from complementary amino acid residues on interacting protein surfaces.
• Predicting these specific interactions remains a challenge.

Purpose of the Study:

• To develop methods for learning and applying probabilistic graphical models of residue cross-coupling constraints.
• To generalize traditional binding motifs with probabilistic semantics for evaluating new interactions.
• To provide explainable predictions of protein-protein interactions.

Main Methods:

• Utilized multiple sequence alignments and known interaction data.
• Developed probabilistic graphical models to capture residue cross-coupling.
• Incorporated techniques to handle various interaction types (one-to-one, many-to-many) and potential biases.

Main Results:

• Identified biologically relevant cross-coupling constraints between protein families.
• Successfully predicted known protein-protein interactions.
• Generated explainable predictions for novel interactions, demonstrated on PDZ domains and their ligands.

Conclusions:

• Probabilistic graphical models offer a powerful framework for understanding and predicting protein-protein interaction specificity.
• The developed approach enhances the evaluation of potential interactions and provides mechanistic insights.
• This method advances the study of protein assemblies and molecular recognition.