Integration of molecular coarse-grained model into geometric representation learning framework for protein-protein...

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning

Background:

• Predicting protein-protein interaction (PPI) properties is crucial for understanding biological processes and developing disease treatments.
• Current structure-based machine learning methods often use atom- or residue-scale representations, which can be computationally intensive and may miss key interaction details.

Purpose of the Study:

• To introduce MCGLPPI, a novel geometric representation learning framework for accurate and efficient prediction of PPI properties.
• To evaluate MCGLPPI's performance against existing atom- and residue-scale methods.

Main Methods:

• MCGLPPI combines graph neural networks (GNNs) with MARTINI molecular coarse-grained (CG) models.
• The framework utilizes CG-scale representations for predicting PPI properties.
• Pre-training on protein domain-domain interaction structures was explored to enhance predictive capabilities.

Main Results:

• MCGLPPI achieved competitive performance on three downstream PPI property prediction tasks compared to atom- and residue-scale methods.
• The CG-scale approach of MCGLPPI consumed only one-third of the computational resources.
• CG-scale pre-training improved the predictive performance for PPI tasks.

Conclusions:

• MCGLPPI provides an effective and efficient solution for predicting PPI properties at the coarse-grained scale.
• This framework offers a promising tool for large-scale biomolecular interaction analysis.
• The study highlights the advantages of coarse-grained representations in machine learning for PPI prediction.