Generative AI-Assisted Discovery of HPK1 Inhibitors

Area of Science:

• Medicinal Chemistry
• Artificial Intelligence
• Drug Discovery

Background:

• Generative artificial intelligence (AI) is increasingly utilized in medicinal chemistry.
• Specific case studies detailing AI applications are emerging in scientific literature.
• Hematopoietic progenitor kinase 1 (HPK1) is a target for inhibitor development.

Purpose of the Study:

• To apply AstraZeneca's REINVENT generative molecular design platform for identifying novel inhibitor scaffolds against HPK1.
• To demonstrate the utility of AI in distinct stages of drug discovery, including hit identification and scaffold hopping.
• To optimize a novel scaffold to achieve potent cellular activity, kinase selectivity, and favorable pharmacokinetics.

Main Methods:

• REINVENT platform utilized transfer learning on kinase-active compounds for hit identification.
• Reinforcement learning with Quantitative Structure-Activity Relationship (QSAR)-based scoring guided initial discovery.
• 3D pharmacophore and docking models served as scoring functions for scaffold hopping.
• Optimization of a lead scaffold was performed to enhance drug-like properties.

Main Results:

• Discovery of three active chemotypes through AI-driven hit identification.
• Identification of two additional active chemotypes via scaffold hopping using 3D pharmacophore and docking.
• Optimization yielded a compound with potent cellular activity, high kinase selectivity, and good rat pharmacokinetics.

Conclusions:

• Generative AI, exemplified by REINVENT, effectively integrates with medicinal chemistry expertise.
• The study validates the application of generative AI for identifying novel inhibitor scaffolds and optimizing drug candidates.
• This approach supports the broader adoption of AI in future drug discovery programs.