Shows how domain skills let a coding agent handle a multi-day cheminformatics loop (data curation, matched molecular pair analysis, candidate proposal) in about an hour. A useful template for packaging expert workflows as agent skills.
Key takeaways · AI-distilled
Google DeepMind's Cheminformatics Skills split the work: the LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → orchestrates the workflow, navigates databases and reasons, while deterministic tools groundingTying a model's answers to checkable sources — retrieved documents, live data, tool results — instead of letting it answer from memory alone.Full definition → in RDKit, with ChEMBL and PubChem connectors, run the molecular calculations.
The walkthrough targets rigidification: adding a bond that ties two substructures together so a molecule pays its entropic binding cost in advance. Lorlatinib resembles crizotinib (five major rotatable bonds) but is cyclized to an effective zero and is more potent.
In the demo, the AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition →'s matched molecular pair analysis compared molecules that differ only in whether they are cyclized, and found that cyclization reduced rotatable bonds while generally increasing potency, as medicinal chemistry predicts.
The skills steer potency modeling toward chronological splits, training on older compounds to predict newer ones when dates are available, and fall back to scaffold splits to better estimate how well a model generalizes.
DeepMind cautions that sparse data, inter-lab assay noise and narrow dynamic ranges limit predictive precision, so the agent's proposed macrocycles are testable hypotheses that must still be synthesized and validated experimentally.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
grounding — Tying a model's answers to checkable sources — retrieved documents, live data, tool results — instead of letting it answer from memory alone.
