Scientists discover hidden binding site in cancer protein that AI tools completely missed

Hidden Drug-Binding Pocket Reveals AI Blind Spots

Researchers at the Icahn School of Medicine at Mount Sinai have uncovered a previously unknown druggable site in PKMYT1, a cancer-related protein that controls how cells grow and divide1

1

. Published in the Journal of the American Chemical Society on June 2, the study demonstrates both the potential and critical limitations of AI tools in modern drug development2

2

. This hidden drug-binding pocket, which current state-of-the-art AI systems completely missed, could provide a new pathway toward developing selective cancer drugs with reduced toxicity.

Why Current Cancer Drug Design Falls Short

Most experimental drugs targeting kinases like PKMYT1 work by blocking the ATP-binding site, the region that uses cellular energy to function1

1

. The challenge lies in specificity: many kinases share nearly identical ATP-binding sites, making it difficult for drugs to distinguish between targets. This lack of selectivity leads to unwanted side effects when drugs inadvertently block other kinases. The newly discovered binding pocket offers an alternative approach that could sidestep these specificity challenges associated with traditional kinase inhibitors.

Experimental Validation Uncovers What AI Missed

The Mount Sinai research team employed AlphaFold2 to predict possible protein structures of PKMYT1, then performed virtual screening to identify potential binding molecules2

2

. However, the hidden pocket only emerged through experimental validation using X-ray crystallography, biochemical testing, and cellular studies. Additional computational methods, including AlphaFold3 and Boltz-2, along with molecular dynamics simulations, were tested afterward—yet none could predict the newly discovered binding mode.

"AI was very accurate when predicting known protein shapes, but it missed a completely unexpected binding pocket that we could only uncover experimentally," says Avner Schlessinger, PhD, Director of the AI Small Molecule Drug Discovery Center at Mount Sinai1

1

. This gap highlights the limitations of AI in drug discovery when proteins exhibit unexpected conformations.

Protein Flexibility Changes the Drug Development Game

The findings reveal that protein flexibility plays a more significant role than previously understood. PKMYT1 constantly shifts between different shapes rather than maintaining a single fixed form2

2

. Even more striking, the study found that tiny chemical modifications to a molecule could dramatically alter binding behavior. "A very small chemical modification caused the molecule to switch from binding in this hidden pocket to binding in a much more conventional way," explains Michael Lazarus, PhD, Associate Professor of Pharmacological Sciences at Mount Sinai1

1

. This sensitivity to subtle molecular changes reinforces why experimental validation in drug discovery remains essential, even as AI drug discovery tools advance.

Implications for Future Selective Therapies

The work could help scientists develop selective cancer drugs that avoid toxicity issues plaguing current kinase inhibitors2

2

. The compounds identified represent promising starting points for further optimization and testing in disease models. Looking ahead, the team plans to develop more potent compounds targeting this site and investigate whether similar hidden pockets exist in other cancer-related kinases. They also aim to refine computational methods so AI systems can better recognize dynamic protein states currently overlooked.

For AI drug discovery platforms, this research serves as a critical reminder: while tools like AlphaFold excel at predicting known protein structures, they struggle with unexpected conformations. The findings may help improve future AI systems by teaching them to recognize hidden and dynamic protein states. As cancer drug design evolves, the Mount Sinai research underscores that combining AI predictions with rigorous experimental validation will be essential for uncovering novel therapeutic opportunities that computational approaches alone cannot detect.