Insilico Medicine's Chemistry42 Designs First-in-Class PKMYT1 Degrader ISM2286

Title

Insilico Medicine's Chemistry42 Designs First-in-Class PKMYT1 Degrader ISM2286

Content

Insilico Medicine used its Chemistry42 generative AI platform to design ISM2286 — a first-in-class PROTAC degrader targeting PKMYT1, a kinase with synthetic lethality to CCNE1-amplified tumors. The AI workflow produced a pyrrolo[2,3-b]pyridine warhead (ISM4963), an 8 Å linker, and a CRBN E3-binding ligand; off-target kinases fell from 46 to 11 — a 76% drop versus RP-6306 — with experimental validation in Nature Communications (Wang et al., 16, 10759, 2025). "The core value of Golden Cubes lies in compressing what traditionally requires weeks of wetlab experimental screening for kinome selectivity into a matter of hours" — Insilico Medicine, Chemistry42 PROTAC design white paper (2026-06-18). Existing PKMYT1 inhibitors like RP-6306 rely on sustained ATP-site occupancy and suffer from poor selectivity, resistance risk, and narrow therapeutic windows. Insilico instead pursued targeted protein degradation: a PROTAC recruits an E3 ligase to trigger ubiquitin-proteasome-mediated PKMYT1 depletion, enabling deeper and longer-lasting pathway suppression than reversible inhibition. Three warhead design requirements guided the work — improved kinase selectivity, a linker-accessible binding orientation, and a PROTAC-compatible physicochemical profile. The project unfolded across three Chemistry42-driven phases. Phase 1 prepared inputs using RP-6306 as a reference ligand with hydrogen-bond pharmacophore constraints, a 5 Å pocket dilation, and a dual-template pharmacophore fusion incorporating dasatinib's solvent-exposed features. Twenty-plus generative AI models proposed scaffolds, triaged by PLI scoring, LE > 0.3 / LLE > 3 reference metrics, 140 ADMET models, and Golden Cubes (a virtual kinome-screening module applying a ≤5-off-target kinase threshold). After medicinal-chemistry refinement added a cyano group to restore a conserved water-mediated hydrogen bond, Compound 3 emerged — but its PK profile was suboptimal (clearance 63.6 mL/min/kg, only 6% oral bioavailability in rats), and Chemistry42's ADMET tools flagged the pyrimidine methyl as an oxidative hotspot. Replacing that methyl with an NH2 tail yielded ISM4963. Phase 2 built a PKMYT1-PROTAC-CRBN-E2 ternary complex model using molecular-dynamics stability screening and AlphaFold3-based E2 prediction, identifying Lys345 as the productive ubiquitination site. Phase 3 used Chemistry42's Anchor Points to lock both terminal binders, then iteratively explored the linker chemical space to discover a piperazine-containing bridge satisfying the 8 Å distance constraint. The platform combined three preset AI generative tiers (Light, Optimal, Advanced), over 900 Medicinal Chemistry Filters, 140 ADMET models, and Golden Cubes. The Golden Cubes predictions were experimentally confirmed — off-target kinases fell 76% (46→11) — and ISM2286 showed favourable cross-species oral bioavailability, validating Chemistry42's capacity to compress conventional PROTAC development by integrating target validation, warhead design, ternary-complex modelling, and linker optimization into a single AI-driven pipeline.