High precision, computationally-guided discovery of highly selective Wee1 inhibitors for the treatment of solid tumors

Wee1 is an attractive oncology target, but the existing inhibitor adavosertib (AZD-1775) had suspected liabilities from off-target PLK-family kinase inhibition and time-dependent CYP3A4 inhibition. In an initial 7-month campaign the team enumerated over 6 million design ideas, filtered and docked them, then ran 9,000 FEP+ calculations evaluating binding affinity and PLK1 selectivity; after discovering further off-target liabilities, a second 3-month campaign used protein FEP+ to profile 6,700 new designs for potency against Wee1 and broad kinome selectivity, supplemented by DeepAutoQSAR machine learning and Jaguar quantum mechanics for ADME, all coordinated through LiveDesign. Multiple chemotypes achieved nanomolar affinity with up to 1000x selectivity for Wee1 over PLK1; compound STC-8123 outperformed AZD-1775 in vivo, and development candidate SGR-3515 -- with superior binding affinity/potency and excellent broad kinome selectivity -- is currently in Phase 1 clinical trials.

The customer is anonymised in Schrödinger’s own case study — not withheld by us.

Schrödinger is a physics-plus-machine-learning computational platform for molecular design; all top-20 pharma companies by revenue are customers of its drug-discovery software.

This sits in the Build stage of the AI use-case lifecycle: Named examples from the foundation-model labs and platforms pharma teams use to actually construct, train and generate the model or application.