Learning Sequence Determinants of Protein:protein Interaction Specificity with Sparse Graphical Models.

Area of Science:

• Structural biology
• Computational biology
• Biophysics

Background:

• Understanding protein-protein interactions is crucial for deciphering biological processes.
• Key questions involve identifying interacting partners, quantifying binding affinity, and elucidating the molecular basis of specificity.
• Large-scale experimental data now allow for the development of predictive models.

Purpose of the Study:

• To develop a method, DgSpi (Data-driven Graphical models of Specificity in Protein:protein Interactions), for predicting the strength (ΔG) and specificity of protein-protein interactions.
• To explicitly represent the amino acid basis for interaction specificity.
• To enable the design of novel interacting protein partners.

Main Methods:

• Development of DgSpi, a graphical model approach.
• Application of DgSpi to predict binding free energy (ΔG) for PDZ domain-peptide interactions.
• Utilizing data from MacBeath and colleagues for 82 PDZ modules and 217 peptide partners.
• Validation using 10-fold cross-validation and leave-one-PDZ-out cross-validation.

Main Results:

• Predicted ΔG values showed high correlation with experimental measurements (r=0.69 in 10-fold CV, r=0.63 in leave-one-PDZ-out CV).
• The model provides residue-level insights into interaction constraints, explaining protein-level predictions.
• DgSpi, as a generative model, successfully designed novel and diverse interacting peptide ligands.

Conclusions:

• DgSpi effectively predicts protein-protein interaction strength and specificity.
• The model elucidates the amino acid determinants of binding specificity.
• DgSpi facilitates the rational design of new protein interaction interfaces.