At the Icahn School of Medicine at Mount Sinai, researchers have uncovered something that artificial intelligence—the technology reshaping drug discovery—simply could not see: a hidden pocket in PKMYT1, a protein that controls how cells grow and divide, where cancer drugs could bind far more precisely than current treatments allow. The discovery, published in the Journal of the American Chemical Society on June 2, reveals both the remarkable promise and the stubborn limitations of AI in the race to develop better cancer medicines.
The problem it solves is a real one. Most experimental cancer drugs targeting kinases like PKMYT1 work by blocking the ATP-binding site—the region where the protein taps into the cell's energy supply. But here's the catch: nearly all kinases share nearly identical ATP-binding sites, making it nearly impossible for drugs to distinguish between the intended target and other kinases in the body. That lack of precision means side effects. It means collateral damage. The quest for cancer drugs that could hit their target without wreaking havoc elsewhere has long been central to oncology.
Enter Avner Schlessinger, Ph.D., a professor of pharmacological sciences and director of the AI Small Molecule Drug Discovery Center at Mount Sinai, along with his colleague Michael Lazarus. The team began using AlphaFold2, one of the most celebrated AI systems for predicting protein structures, to map out possible shapes PKMYT1 might take. They screened for molecules that could interact with it, then moved into the laboratory with X-ray crystallography, biochemical tests, and cellular studies to validate what the models predicted. The experimental work revealed what the AI had missed entirely: an entirely new binding pocket, a completely hidden site where drug molecules could attach.
The implications ripple outward. If drugs could be designed to bind to this hidden pocket instead of the crowded ATP-binding site, they could potentially be far more selective—hitting PKMYT1 while leaving other kinases untouched. Fewer side effects. More precise medicine. But the story doesn't end with celebration of AI's prediction power.
What the team discovered next humbles the technology even further. Using AlphaFold3 and Boltz-2, along with molecular dynamics simulations, they tested whether current computational tools could have predicted this newly discovered binding mode. They couldn't. And there's more: a tiny chemical modification—a small tweak to a molecule's structure—caused it to switch from binding in the hidden pocket to binding in the conventional way. The proteins, in other words, are far more dynamic and sensitive than anyone had appreciated. They don't sit still. They shift and reshape themselves in response to subtle changes.
"That tells us these proteins are incredibly dynamic and sensitive to subtle molecular changes," Lazarus noted. "It also reinforces why experimental validation remains essential, even in the era of AI."
For cancer patients, the path forward is clearer because of human ingenuity working in tandem with machine learning—not replacing it. The hidden pocket discovered at Mount Sinai may eventually lead to a new generation of cancer drugs that are both more effective and less toxic. But the research also suggests that the future of drug discovery depends not on letting AI run alone, but on keeping humans in the laboratory, asking questions, running experiments, and catching what the algorithms miss.