Wee1 inhibitor optimization through deep-learning-driven decision making.

Area of Science:

• Computational chemistry and drug discovery
• Artificial intelligence in medicinal chemistry
• Oncology and cancer therapeutics

Background:

• Deep learning (DL) shows promise in accelerating hit screening and molecular optimization.
• Wee1 kinase is a validated target for cancer therapy, with inhibitors showing significant potential.
• Optimizing lead compounds requires efficient and accurate predictive methodologies.

Purpose of the Study:

• To apply a multi-technique deep learning strategy for the optimization of Wee1 inhibitors.
• To generate novel Wee1 inhibitors with enhanced potency and anti-cancer activity.
• To demonstrate the utility of DL in streamlining the molecular optimization process.

Main Methods:

• Utilized a deep learning pipeline encompassing activity interpretation, scaffold-based molecular generation, and activity prediction.
• Optimized the in-house Wee1 inhibitor GLX0198 using the developed DL strategy.
• Synthesized and tested generated compounds for Wee1 inhibitory activity and cancer cell line efficacy.

Main Results:

• Generated three optimized Wee1 inhibitors with significantly improved potency (IC50 values ranging from 13.5 nM to 47.1 nM) from five selected molecules.
• Identified highly potent Wee1 inhibitors through further minor modifications, demonstrating desirable inhibitory effects across multiple cancer cell lines.
• The best compound (13) exhibited superior cancer cell inhibition, with IC50 values below 100 nM in all tested cancer cell lines.

Conclusions:

• The integrated deep learning approach effectively facilitated molecular optimization of Wee1 inhibitors.
• Deep learning strategies can significantly accelerate the identification of potent drug candidates for cancer therapy.
• This study highlights the potential of AI in advancing medicinal chemistry and drug discovery pipelines.