ChemScreener: an active learning enabled hit discovery workflow with WDR5 inhibitor case study.

Area of Science:

• Drug Discovery
• Computational Chemistry
• Machine Learning

Background:

• Hit identification is a critical yet costly and time-consuming phase in drug discovery.
• Active learning strategies offer potential for optimizing screening efficiency with limited data.
• Exploring chemical space and adapting to data scarcity remain key challenges in active learning for drug discovery.

Purpose of the Study:

• To present ChemScreener, a multi-task active learning workflow for early drug discovery.
• To evaluate ChemScreener's Balanced-Ranking acquisition strategy for exploring novel chemistry and enriching hit rates.
• To demonstrate ChemScreener's effectiveness in accelerating hit discovery and identifying diverse chemotypes.

Main Methods:

• Development of ChemScreener, a multi-task active learning workflow.
• Implementation of a Balanced-Ranking acquisition strategy leveraging ensemble uncertainty.
• Iterative high-throughput screening (HTS) using time-resolved fluorescence resonance energy transfer (TR-FRET) assays on WDR5 protein.
• Consolidation, retesting, clustering, dose-response analysis, and differential scanning fluorimetry (DSF) validation of identified hits.

Main Results:

• ChemScreener increased hit rates from 0.49% in primary HTS to an average of 5.91% (104 hits from 1760 compounds) over five iterative screens.
• 44 hit compounds (from 81 clusters) advanced to dose-response assays, with over 50% validating as WDR5 binders (IC50 < 45 μM) by DSF.
• Three novel scaffold series and three singleton scaffolds were identified de novo.

Conclusions:

• ChemScreener effectively accelerates early hit discovery and enhances hit rate enrichment.
• The workflow successfully identifies diverse chemotypes, expanding chemical space exploration.
• ChemScreener provides a generalizable, scalable, and effective framework for ligand-based virtual screening in drug discovery.