Google DeepMind adds chemistry skills to Antigravity

Google DeepMind described a suite of Cheminformatics Skills for Google Antigravity. The company says the skills turn Antigravity into "a collaborative science workbench" with full access to local tooling, for both computational chemists and bench chemists.

The design splits the work in two. The language model handles workflow orchestration, database navigation and scientific reasoning. Deterministic chemistry tools run the molecular calculations. The skills are built on RDKit, the open-source cheminformatics toolkit developed by a community of contributors over two decades. They also include connectors to curated data from ChEMBL and PubChem. According to the post, the skills build domain best practices directly into the agent's workflow.

The problem being addressed is data wrangling. The same molecule can appear as an IUPAC name in a paper, a Markush structure in a patent, or an SDF file in a vendor catalog. Errors here are costly: change a single atom or bond and you are describing a different compound.

The post's walkthrough uses rigidification, a standard medicinal chemistry technique. Chemists add bonds to cut down a molecule's conformational freedom and reduce the entropic penalty it pays when binding. The example compares the lung cancer drugs crizotinib and lorlatinib. Crizotinib, a first-generation kinase inhibitor, has five major rotatable bonds. Lorlatinib, a third-generation inhibitor, is cyclized so its effective number of rotatable bonds drops to zero, which contributes to its higher potency.

The excerpt does not report benchmarks, accuracy figures or availability details. It also does not show the agent's own output from the walkthrough.