MO-MEMES: A method for accelerating virtual screening using multi-objective Bayesian optimization.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Identifying drug candidates requires screening large libraries for multiple properties like binding affinity and LogP.
• High-throughput screening is time-consuming and computationally expensive for vast molecular databases.
• Existing methods struggle to efficiently evaluate numerous physical properties simultaneously.

Purpose of the Study:

• To develop an advanced machine learning framework for efficient multi-objective drug screening.
• To reduce the computational burden of evaluating physical properties for large drug libraries.
• To enhance the identification of drug-like molecules with desired characteristics.

Main Methods:

• Extension of the Machine learning framework for Enhanced MolEcular Screening (MEMES).
• Application of multi-objective Bayesian optimization for molecular property prediction.
• Focused property calculation on a small subset of the drug library.

Main Results:

• Identified over 90% of the most desirable molecules based on multiple required properties.
• Achieved this by explicitly calculating properties for only 6% of the drug library.
• Demonstrated significant efficiency gains in hit identification.

Conclusions:

• The enhanced MEMES framework offers a powerful solution for accelerating drug discovery.
• Multi-objective Bayesian optimization drastically improves screening efficiency.
• This approach facilitates the identification of drug candidates with a comprehensive set of desired properties.