• Sources: primary
  • Summary: The skills let the model orchestrate while deterministic RDKit tools do molecular calculations, with connectors to ChEMBL and PubChem. In a crizotinib to lorlatinib rigidification walkthrough the agent used chronological or scaffold splits for potency models and proposed synthesis routes in about an hour. DeepMind says designs remain hypotheses that need experimental validation.
  • Why it matters: It is a concrete pattern for drug discovery agents that keep chemistry computation in deterministic tooling rather than in the model.