Ollivier Persistent Ricci Curvature-Based Machine Learning for the Protein-Ligand Binding Affinity Prediction

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Efficient molecular featurization is crucial for machine learning in drug design.
• Existing methods often struggle with capturing complex molecular structures.

Purpose of the Study:

• To introduce Persistent Ricci Curvature (PRC), specifically Ollivier PRC (OPRC), as a novel molecular descriptor.
• To develop and evaluate an OPRC-based machine learning model for predicting protein-ligand binding affinity.

Main Methods:

• Generated nested molecular graphs using persistent homology.
• Defined and calculated Ollivier Ricci curvatures (OPRC) on these graphs.
• Extracted persistent attributes as molecular descriptors.
• Integrated OPRC descriptors with a gradient boosting tree (GBT) model.

Main Results:

• The OPRC-GBT model achieved state-of-the-art performance in predicting protein-ligand binding affinity.
• Demonstrated advantages over traditional molecular descriptors on PDBbind datasets.
• Showcased the effectiveness of PRC for molecular featurization.

Conclusions:

• OPRC offers a powerful new method for molecular featurization and feature engineering.
• The OPRC-GBT model represents a significant advancement in drug design, particularly for affinity prediction.
• This approach holds promise for improving the efficiency and accuracy of drug discovery pipelines.