Enabling Open Machine Learning of Deoxyribonucleic Acid-Encoded Library Selections to Accelerate the Discovery of...

Area of Science:

• Drug discovery and development
• Computational chemistry
• Bioinformatics

Background:

• Machine learning (ML) is vital for DNA-encoded library (DEL) screening in ligand discovery.
• Proprietary datasets limit accessibility and reproducibility in DEL-ML applications.
• There is a need for open and automated frameworks to democratize drug discovery.

Purpose of the Study:

• To present the first fully open, automated DEL-ML framework.
• To enable reproducible and accessible drug discovery using public DEL data.
• To demonstrate the efficacy of open-source approaches in ligand discovery.

Main Methods:

• Developed an automated workflow for ML model training, virtual screening, and compound selection.
• Utilized public DEL datasets and chemical fingerprints for model development.
• Applied the framework to identify WDR91 binders using large-scale virtual screening.

Main Results:

• Identified 50 candidate compounds for WDR91.
• Experimentally validated seven novel binders with dissociation constants in the micromolar range (2.7–21 μM).
• The open-source framework demonstrated performance comparable to proprietary methods.

Conclusions:

• Public DEL data can support robust ML-driven ligand discovery.
• The open-source DEL-ML framework enhances transparency and accessibility in drug development.
• This approach fosters broader community participation in discovering novel therapeutics.