Machine Learning on DNA-Encoded Libraries: A New Paradigm for Hit Finding.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• DNA-encoded small molecule libraries (DELs) are powerful tools for identifying novel inhibitors against therapeutic targets.
• Traditional DEL screening can be resource-intensive, necessitating innovative approaches for hit identification.

Purpose of the Study:

• To develop and validate a machine learning-based approach for hit-finding using DEL selection data.
• To identify active molecules from large compound libraries, including commercial and synthesizable compounds.

Main Methods:

• Training machine learning models exclusively on DEL selection data.
• Applying automated filters to model predictions for hit prioritization.
• Conducting a prospective study of approximately 2000 compounds across three targets: sEH, ERα, and c-KIT.

Main Results:

• Achieved an overall hit rate of approximately 30% at 30 μM across all tested targets.
• Discovered potent compounds with IC50 values below 10 nM for each target.
• Identified diverse, drug-like compounds that are structurally distinct from known ligands.

Conclusions:

• Machine learning applied to DEL selection data offers a powerful and efficient strategy for hit-finding.
• This method successfully identifies potent inhibitors for diverse protein targets, even for molecules dissimilar to the original DEL library.
• The approach accelerates the discovery of novel, drug-like chemical matter for therapeutic development.