Insilico Uses AI Chemistry to Discover Selective PKMYT1 Inhibitors
Insilico Medicine applied its generative chemistry platform Chemistry42 to design novel small molecules and PROTACs that achieve subfamily selectivity against the serine/threonine kinase PKMYT1, a validated synthetic-lethal oncology target. The team replaced the pyrido‑pyrrole core of clinical lead RP-6306 (RE1) with a thiazolyl‑pyrazole scaffold that forms an internal sulfur–lone-pair interaction to lock a syn‑coplanar conformation. Lead A4-ent1 shows high biochemical potency (IC50 2.2 nM) and substantially improved selectivity margins versus off-targets like BRAF, RAF1, and SRC, addressing dose-limiting toxicities seen with prior molecules. The work — highlighted as a ChemMedChem cover feature — demonstrates how targeted conformational design guided by generative AI can resolve kinase selectivity trade-offs important for translational medicinal chemistry.
What happened
Insilico Medicine used its generative chemistry suite Chemistry42 to design a new series of selective inhibitors and PROTAC degraders against the serine/threonine kinase PKMYT1, a clinically validated synthetic-lethal target for cancers with CCNE1 amplification. The team substituted the pyrido‑pyrrole core of clinical comparator RP-6306 (RE1) with a thiazolyl‑pyrazole scaffold that exploits a noncovalent sulfur–lone-pair interaction to enforce a rigid, syn‑locked coplanar conformation. The best enantiomer, A4-ent1, demonstrates biochemical potency at IC50 2.2 nM and markedly improved selectivity versus off-target kinases (BRAF, RAF1, SRC).
Technical details
The approach emphasizes conformational restriction and internal heteroatom interactions rather than classical hydrogen-bond strategies. Key practical takeaways for practitioners:
- •The scaffold swap to a thiazolyl‑pyrazole system leverages an internal sulfur–nitrogen lone-pair interaction to lock geometry.
- •Chemistry42 guided generation prioritized physicochemical masking of hydrogen-bond donors while maintaining optimal active-site geometry.
- •Outputs included both small-molecule inhibitors and highly selective PROTAC designs intended to reduce systemic exposure and off-target toxicity.
Context and significance
Kinase ATP-binding sites are highly conserved, making subfamily-level selectivity a persistent medicinal chemistry challenge. Clinical leads such as RP-6306 have demonstrated proof-of-concept but suffer narrow selectivity windows (<10x) that cause dose-limiting adverse events. This work shows that AI-driven generative design can surface non-obvious intramolecular interactions and scaffold hops that improve both potency and selectivity, accelerating hypothesis generation and lead optimization cycles. For computational chemistry teams, the study underlines the practical value of integrating generative models with explicit conformational and heteroatom interaction hypotheses rather than treating generation as a purely ligand-centric scoring exercise.
What to watch
Validate cellular efficacy, in vivo tolerability, and whether the selectivity improvements translate to widened therapeutic windows in animal models. Also watch independent replication of the internal lone-pair design pattern across other kinase families.
Scoring Rationale
The result is directly relevant to computational chemists and translational teams: it demonstrates a pragmatic use of generative AI to solve a longstanding kinase selectivity problem. It is notable but not industry-defining; pending in vivo and clinical translation, the immediate practical impact is moderate-high. (Recent publication date subtracts a small freshness penalty.)
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problemsStep-by-step roadmaps from zero to job-ready — curated courses, salary data, and the exact learning order that gets you hired.
