Persistent Path-Spectral (PPS) Based Machine Learning for Protein-Ligand Binding Affinity Prediction.

Area of Science:

• Computational Chemistry
• Cheminformatics
• Machine Learning

Background:

• Molecular descriptors are crucial for quantitative structure-activity/property relationship (QSAR/QSPR) and machine learning models.
• Accurate prediction of protein-ligand binding affinity is vital in drug discovery.

Purpose of the Study:

• To propose persistent path-spectral (PPS) molecular descriptors and a machine learning model for predicting protein-ligand binding affinity.
• To introduce a novel topological approach for molecular characterization.

Main Methods:

• Constructing path-Laplacian from graph, simplicial complex, and hypergraph representations of molecules.
• Deriving persistent path-spectral from a filtration process for multiscale molecular characterization.
• Developing a PPS-based machine learning model (PPS-ML) using gradient boosting trees.

Main Results:

• The PPS-based molecular descriptors provide quantitative descriptions of molecules.
• The PPS-ML model achieved competitive performance on PDBbind-v2007, PDBbind-v2013, and PDBbind-v2016 datasets.
• Demonstrated the efficacy of topological data analysis in predicting binding affinity.

Conclusions:

• Persistent path-spectral descriptors offer a novel and effective way to represent molecular structures.
• The PPS-ML model shows promise for accurate protein-ligand binding affinity prediction.
• This approach advances the application of topological methods in cheminformatics.