Researchers at the Icahn School of Medicine at Mount Sinai have identified a previously unknown “hidden” binding site in a protein associated with cancer development. The discovery could pave the way for a new generation of more precise anticancer drugs while also highlighting the current limitations of artificial intelligence in drug discovery.

The study, published in the Journal of the American Chemical Society, focused on PKMYT1, a kinase involved in regulating cell growth and division. Because disruptions in kinase activity are frequently linked to cancer, PKMYT1 has emerged as a promising therapeutic target.

Most experimental kinase inhibitors act on the protein’s ATP-binding site—the region where the enzyme interacts with the energy-carrying molecule ATP. However, this site is highly similar across many different kinases, making it difficult to develop drugs that target only a specific protein without affecting others and causing unwanted side effects.

Using a combination of AI-based protein structure prediction tools and laboratory techniques, the researchers discovered an alternative, previously hidden pocket within PKMYT1. This cavity allows molecules to bind in a manner distinct from the protein’s conventional active site.

Remarkably, modern artificial intelligence systems, including AlphaFold2, failed to predict the existence of this binding pocket despite their impressive accuracy in modeling known protein structures.

To validate their findings, the team employed X-ray crystallography, biochemical assays, and cell-based experiments. Additional computational approaches—including AlphaFold3, Boltz-2, and molecular dynamics simulations—were also used to test and refine the binding models.

According to Avner Schlessinger, the findings demonstrate that proteins are far more dynamic than previously appreciated. Rather than existing in a single fixed shape, they constantly shift between different conformations. Even subtle changes in a molecule’s chemical structure were enough to switch binding between the newly discovered hidden pocket and the more traditional binding site.

Co-author Michael Lazarus noted that these subtle structural differences underscore the importance of experimental validation, even when using the most advanced AI systems available today.

The discovery could ultimately contribute to the development of more selective cancer therapies by providing drug designers with a new target that may offer greater specificity than conventional ATP-binding sites. It also suggests that future AI models may need to better account for hidden and highly dynamic protein states, which can be difficult to capture using current prediction methods.

More broadly, the study serves as a reminder that while AI has become a powerful tool in biomedical research, experimental science remains essential for uncovering biological features that may not be apparent from computational models alone.