ToolsNews
Antigravity adds Cheminformatics Skills for drug design
Google DeepMind released Cheminformatics Skills for Antigravity. In its demo, the agent goes from literature search to novel macrocycle designs in about an hour, a loop that used to take days.
By: Mikola Máté Patrik · · 2 min read
Mikola Máté Patrik
What happened?
Google DeepMind released Cheminformatics Skills for Google Antigravity, its agentic development environment (Google DeepMind). The skills install through the Science plugin in the Antigravity app, and Enterprise users can also access them with Google Cloud credentials.
DeepMind's demo uses lung cancer drugs: a workflow that typically took several days per iteration runs in roughly an hour. The agent pulls thousands of compounds from the literature, analyzes them, and proposes new molecules along with synthetic routes.
How does it work?
The split is clean: the language model handles orchestration, database navigation and scientific reasoning, while deterministic chemistry tools run the molecular calculations. The base is the open source RDKit toolkit, with connectors to ChEMBL and PubChem, and the skills encode domain best practices directly into the agent's workflow.
The example is rigidification. Lorlatinib resembles crizotinib but is cyclized, which brings its effective number of rotatable bonds to zero and raises potency. The agent collects analogues of both drugs, filters out buffers and solvents, launches an interactive viewer, runs matched molecular pair analysis (MMPA), and trains and validates several potency models using scaffold or chronological splits.
Finally, it proposes new rigidified macrocycles with synthetic routes and curved arrow electron pushing diagrams, scoring each design with the potency models it trained earlier. Results export to industry standard chemistry software, and designs can go on to AlphaFold or molecular dynamics.
Where are the limits?
DeepMind itself flags that sparse data, inter-lab assay noise and narrow dynamic ranges limit predictive precision. The computational designs are testable hypotheses, not finished drugs, and every proposal still has to be synthesized and validated experimentally.
What does it mean in practice?
Antigravity is stretching from a coding environment into a science workbench, and the pattern generalizes: the LLM orchestrates, deterministic tools do the exact math. If you're building an agent for your own domain, that division of labour is worth copying.
Google is offering 10,000 scientists the highest Antigravity usage tier free for six months. Applications are open, subject to eligibility, capacity and selection.
Produced with the help of AI, reviewed by an editor. How we make our articles
Reviewed and edited by the Neural Daily desk. About
