Active Learning-Guided Hit Optimization for the Leucine-Rich Repeat Kinase 2 WDR Domain Based on In Silico...

Area of Science:

• Computational chemistry and drug discovery
• Neuroscience and neurodegenerative diseases
• Machine learning applications in pharmacology

Background:

• Leucine-rich repeat kinase 2 (LRRK2) mutations are the leading cause of familial Parkinson's disease.
• The LRRK2 WDR domain is an underexplored but crucial drug target for Parkinson's disease.
• No inhibitors for the LRRK2 WDR domain were known before the CACHE Challenge.

Purpose of the Study:

• To design and experimentally validate novel LRRK2 WDR domain inhibitors.
• To apply and assess an active learning (AL) machine learning (ML) workflow for drug discovery.
• To explore chemical spaces efficiently for small-molecule analogs with enhanced binding affinity.

Main Methods:

• Utilized an active learning (AL) machine learning (ML) workflow.
• Employed optimized free-energy molecular dynamics (MD) simulations with the thermodynamic integration (TI) framework.
• Expanded chemical series around confirmed hit molecules for inhibitor design.

Main Results:

• Identified 8 novel LRRK2 WDR domain inhibitors.
• Achieved a 23% hit rate from 35 experimentally tested molecules.
• Demonstrated the efficiency of the free-energy-based AL/ML workflow in exploring chemical spaces.

Conclusions:

• The developed free-energy-based active learning workflow is effective for rapid and efficient exploration of chemical spaces.
• This workflow significantly accelerates the discovery of small-molecule inhibitors with increased affinity.
• The methodology is broadly applicable to screening chemical spaces for drug discovery, with a mean absolute error of 2.69 kcal/mol for TI MD calculations.