Generative AI Designs Spirocyclic Dual Degraders Bridging Innate and Adaptive Immunity Kinases
- A VAE-RNN generative model in the POLYGON framework explored chemical space between two unrelated kinase families, IRAK4 and TYK2.
- Iterative sampling, scoring and refocusing cycles avoided the limits of simple pharmacophore merging and produced a structurally novel spirocyclic scaffold.
- Water-site analysis with WaterMap guided optimization, and PROTAC conjugation then gave a dual degrader, C-8-014.
- The compound efficiently degrades both kinases, supporting a closed-loop AI design approach for hard-to-reach kinome regions.
💡 Points Based on the abstract
Spirocyclic cores are attractive for 3D character and IP space, but a dual IRAK4/TYK2 binder is only half the story here. The ATP pockets share little beyond the hinge, so I'd want biochemical affinity and kinome selectivity for the parent warhead before PROTAC conjugation, plus a non-degrading control. Degradation can also hide weak binding, so ternary complex and E3 dependence need checking, and so does synthetic tractability of the spirocycle.
📄 Paper · Journal of Medicinal Chemistry (ACS) ↗ Generative AI-Driven Interpolation of Cross-Family Kinase Space: Discovery of Dual-Targeting Degraders of IRAK4 and TYK2