Google DeepMind introduced Cheminformatics Skills for Antigravity on November 3, extending the agent development app into a workbench for computational and experimental chemists. The skills help researchers collect molecular data, compare related compounds, build predictive models and propose designs for further investigation.
The language model coordinates the workflow, while deterministic chemistry tools perform molecular calculations. The suite uses RDKit and connections to resources including ChEMBL and PubChem, helping connect the different representations of a molecule found in literature and databases.
A Research Workflow, Not a Finished Medicine
Google's walkthrough follows compounds related to lung cancer treatments, using molecular rigidification as its example. Reducing a molecule's freedom to change shape can help it bind a target. The demonstration gathers compounds, compares pairs that differ by a structural change, trains potency models and suggests new macrocycles and possible synthesis routes.
Google says this demonstrated sequence took about an hour, compared with a workflow that typically takes days. That is a company demonstration, not a general speed guarantee or evidence that a new medicine has been validated.
The article describes testing predictive models on held-out data and using scaffold splits, which separate groups of molecular structures, to assess generalization. Sparse data, differences between laboratories and limited measurement ranges still constrain predictions. Every proposed design needs synthesis and experimental validation.
Access and the Scientist Offer
The skills are available through Antigravity's Science plugin. Enterprise users can choose Use Business Account and sign in with Google Cloud credentials. The announcement does not specify a separate price for these skills.
Google is also accepting applications from scientists for six months on Antigravity's highest usage tier. The offer covers 10,000 scientists, with rolling onboarding subject to eligibility, capacity and selection. It is an application program rather than automatic free access for every researcher.