Targeting WEE1 kinase: an integrated machine learning-cheminformatics framework for ultra-large-scale virtual...

Area of Science:

• Oncology
• Medicinal Chemistry
• Computational Drug Discovery

Background:

• WEE1 kinase is a therapeutic target in cancers with DNA damage response (DDR) defects.
• Existing WEE1 inhibitors face challenges like toxicity and resistance, necessitating new chemotypes.
• Expanding structural diversity is crucial for developing effective WEE1-targeting drugs.

Purpose of the Study:

• To establish a scalable, machine learning-integrated virtual screening framework for ultra-large chemical spaces.
• To identify novel, structurally diverse WEE1 kinase inhibitors.
• To provide starting points for WEE1-targeted oncology drug discovery.

Main Methods:

• Employed a machine learning framework integrating ECFP4 fingerprints, UMAP, and K-means clustering for diversity-guided screening of ~884 million compounds.
• Utilized multi-stage structure-based computational evaluations: pharmacophore modeling, molecular docking, MM/GBSA rescoring, and molecular dynamics simulations.
• Conducted in silico selectivity profiling against PLK1 and preliminary in vitro testing of promising candidates.

Main Results:

• Identified 18 high-confidence WEE1 inhibitor candidates from ultra-large virtual screening.
• Selected three structurally novel scaffolds with predicted stable binding modes and favorable developability profiles.
• Demonstrated potential for selectivity against PLK1 and preliminary antiproliferative activity in triple-negative breast cancer (TNBC) models.

Conclusions:

• The developed computational framework enables efficient hit identification from ultra-large chemical libraries.
• Reported novel scaffolds represent promising starting points for WEE1-targeted cancer therapy.
• The study provides a computational basis for achieving WEE1 inhibitor selectivity.