Active Learning Guided Drug Design Lead Optimization Based on Relative Binding Free Energy Modeling

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Accurate prediction of ligand binding free energy (BFE) is crucial for identifying potent protein inhibitors.
• Traditional methods like thermodynamics integration (TI) using molecular dynamics (MD) simulations are accurate but computationally expensive and time-consuming.

Purpose of the Study:

• To develop an efficient automated workflow for identifying compounds with the lowest BFE from large ligand sets.
• To accelerate the drug discovery process by reducing the computational cost of BFE calculations.

Main Methods:

• Implemented an active learning (AL) and automated machine learning (AutoML) workflow (AL-AutoML).
• Utilized AL-AutoML to efficiently search and select a small subset of high-performing molecules requiring fewer TI calculations.
• Applied the workflow to identify inhibitors for SARS-CoV-2 papain-like protease.

Main Results:

• Identified 133 compounds with improved binding affinity against SARS-CoV-2 papain-like protease.
• Discovered 16 compounds exhibiting over 100-fold improvement in binding affinity.
• Achieved a hit rate superior to traditional expert-driven medicinal chemistry campaigns.
• Demonstrated at least a 20× speedup compared to brute-force approaches.

Conclusions:

• The AL-AutoML workflow combined with free energy simulations offers a highly efficient approach for identifying potent protein inhibitors.
• This method significantly accelerates drug discovery by reducing computational demands.
• The workflow provides a powerful tool for selecting promising drug candidates with enhanced binding affinity.