RASPD+: Fast Protein-Ligand Binding Free Energy Prediction Using Simplified Physicochemical Features

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning applications in pharmacology

Background:

• Virtual screening is crucial for identifying drug candidates.
• Traditional docking methods generate many poses, requiring extensive computation for binding affinity estimation.
• There is a need for faster, more efficient ligand prioritization methods in drug discovery.

Purpose of the Study:

• Introduce Rapid Screening with Physicochemical Descriptors + machine learning (RASPD+) as a novel pre-filtering method.
• Improve ligand prioritization accuracy and speed in drug discovery workflows.
• Utilize pose-invariant physicochemical descriptors for enhanced computational efficiency.

Main Methods:

• Developed RASPD+, a machine learning-based pre-filtering method.
• Employed pose-invariant physicochemical descriptors of ligands and protein binding pockets.
• Trained the models on the PDBbind dataset for regression analysis.
• Evaluated performance against existing methods and traditional scoring functions.

Main Results:

• RASPD+ demonstrates superior regression performance compared to the original RASPD method and traditional scoring functions.
• The method achieves high accuracy without the need for generating ligand poses.
• Identified key molecular features contributing to protein-ligand binding affinity.
• Successfully enriched active molecules from decoy sets, showcasing its practical utility.

Conclusions:

• RASPD+ offers a computationally efficient and accurate approach for ligand prioritization in drug discovery.
• The method accelerates the identification of potential drug candidates by bypassing complex pose generation.
• RASPD+ provides insights into molecular features governing binding affinity, aiding in rational drug design.