Karen Akinsanya

Leader at Schrödinger. Presented at Session I on the dynamic proteome and the ceiling of structure-based drug design.

Key Arguments

We’ve built a trillion-dollar industry on ~900 protein targets; there are ~20,000 proteins and millions of proteoforms
Static structures are necessary, not sufficient — cryptic pockets and rotamer dynamics routinely break clean SAR stories
The SARM1 case study: a cryptic pocket only visible once a molecule engaged, and a dose-dependent biomarker reversal that classical gates missed
Network biology is mostly unexplored — ~92% of the PPI network is empty, only ~4% of estimated 650,000 PPIs have resolved structures
The training data for “dynamic biology” mostly doesn’t exist yet — whoever builds those datasets owns the next decade

“We can’t leverage AI/ML when the data isn’t in the dataset”
Public chemistry and structure data are saturated — the next wave of value comes from generating data that isn’t yet in the training set
Flagged a paper showing an ML model learned which hospital a slide came from, not the underlying biology

Physics-discovered compound, asset to clinic-ready in 10 months. 100% response rate in Waldenström’s macroglobulinemia.

“Robustness, as Kitano put it, is the feedback that maintains homeostasis in health. In cancer, that same robustness defends a deadly equilibrium.”