A Real World Example: Using AI Tools for Drug Discovery

Fri, 21 Aug 2026 00:00:00 +0000

While AI is a hot topic in drug discovery, a growing body of research is demonstrating its practical impact in the field. In a newly published study in the Journal of Medicinal Chemistry, researchers at AstraZeneca shared their progress using AI to discover novel HPK1 inhibitors. Alongside their traditional pipeline, the team deployed the generative AI tool REINVENT to identify novel chemical scaffolds with activity against the HPK1 kinase.

To train REINVENT, the researchers relied on a dataset of over 60,000 compounds. This setup highlights a major bottleneck in modern drug discovery: generating such extensive data against a single target is currently only realistic for established pharmaceutical companies. Very few academic research groups possess datasets of this scale, underscoring how open-sourcing high-quality data could fundamentally accelerate global drug discovery projects.

Using this training data, REINVENT generated several novel scaffolds that were subsequently synthesized in the lab. Critically, the authors emphasized that AI design cannot be followed blindly. Generative outputs must be combined with traditional medicinal chemistry strategies, alongside assessments of metabolic stability and synthetic accessibility. This balanced approach proved its worth: while one generated structure was a rediscovery from the training set, two scaffolds were completely new, validating the real-world utility of generative models.

The team also demonstrated REINVENT’s capability for scaffold hopping, diversifying the chemical space surrounding the initial hit compound. The authors highlighted this step as a vital risk-mitigation strategy, delivering viable backup scaffolds should the primary candidate fail in clinical trials.

Ultimately, this case study offers a clear blueprint for integrating generative AI into drug discovery. While these models offer powerful structural ideation, they cannot replace human expertise; they must be continuously guided, filtered, and interpreted by experienced medicinal chemists. Success in AI-driven discovery will not rely on algorithms alone, but on the powerful combination of rich datasets, advanced modeling, and expert critical judgment.

Read more: Generative AI-Assisted Discovery of HPK1 Inhibitors Journal of Medicinal Chemistry 2026

Artificial Intelligence |

A Real World Example: Using AI Tools for Drug Discovery