Classifying Protein-Protein Binding Affinity with Free-Energy Calculations and Machine Learning Approaches

Area of Science:

• Neuroscience
• Computational Biology
• Biophysics

Background:

• Neuronal wiring in the fruit fly, Drosophila melanogaster, is crucial for brain function.
• The Dpr-DIP interactome is implicated in neuronal wiring, but only a subset of complexes show strong binding affinity.

Purpose of the Study:

• To elucidate the residue-level molecular basis for differential binding affinities within the Dpr-DIP interactome.
• To develop and validate a predictive computational model for protein-protein binding affinity.

Main Methods:

• Binding free-energy calculations using statistical mechanics simulations and a geometrical route.
• Application of machine learning algorithms, including linear discriminant analysis and random forest.
• Cross-species validation of the predictive model on similar protein families.

Main Results:

• Accurate reproduction of experimental binding affinities for two complexes.
• Prediction of binding free energy for two low-affinity complexes and identification of key residues.
• Machine learning model achieved high accuracy (0.99) in distinguishing strong from weak binders.
• Model demonstrated robustness and reliability across 13 diverse species.

Conclusions:

• Computational approaches, combining physics-based simulations and machine learning, can effectively decipher the molecular determinants of protein-protein binding affinity.
• The developed machine learning model offers a broadly applicable tool for interactome analysis and identification of critical binding residues.
• The findings highlight the potential for cross-species prediction of protein interactions, advancing our understanding of conserved biological mechanisms.