Comparative analysis of machine learning methods in ligand-based virtual screening of large compound libraries

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Ligand-based virtual screening (LBVS) is crucial for identifying drug leads.
• Machine learning (ML) methods are increasingly applied to LBVS.
• ML models predict compound properties using structural and physicochemical data.

Purpose of the Study:

• To review advancements in ML for virtual screening of pharmacodynamically active compounds.
• To compare ML tools with traditional structure-based and ligand-based methods.
• To discuss improving ML performance for large-scale library screening.

Main Methods:

• Literature review of ML applications in LBVS.
• Comparative analysis of ML, structure-based, and pharmacophore/clustering methods.
• Evaluation of ML capabilities for diverse chemical structures and complex structure-activity relationships.

Main Results:

• ML methods show promise in predicting compound properties without 3D target structures.
• ML performance is comparable to other LBVS techniques.
• ML offers advantages in handling complex structure-activity relationships.

Conclusions:

• ML is a valuable tool for ligand-based virtual screening in drug discovery.
• Further research can optimize ML for screening large compound libraries.
• ML complements existing virtual screening strategies.