Google DeepMind has turned its Antigravity agent into a drug-discovery workbench, adding Cheminformatics Skills that let it search thousands of compounds, analyse them, and propose new drug candidates with synthesis routes in about an hour. The walkthrough, published on 3 November, uses real lung-cancer drugs to show the workflow, and the company is offering 10,000 scientists its highest Antigravity usage tier free for six months.
What DeepMind built
The new Cheminformatics Skills turn Antigravity into what DeepMind calls a collaborative science workbench, with full access to local tooling for computational and experimental chemists alike. The language model handles orchestration, database navigation and scientific reasoning, while deterministic chemistry tools run the molecular calculations, so the agent follows domain best practice instead of improvising its own chemistry.
Underneath, the skills rest on RDKit, the open-source cheminformatics toolkit built over two decades by a global community, plus connectors to ChEMBL and PubChem, two curated public chemical resources. That grounding matters because chemistry has no margin for error: change a single atom or bond and the result is a different compound, which can turn a promising inhibitor inert.
The lung-cancer walkthrough
DeepMind demonstrates the system on a classic medicinal-chemistry problem: rigidification, the technique of locking a floppy molecule into the shape that binds its target. The walkthrough starts with crizotinib, a first-generation lung-cancer drug with five major rotatable bonds, and its cyclised successor lorlatinib, which has effectively none and binds more potently.
Asked to search the literature for those drugs and their analogues, the agent retrieves thousands of compounds within minutes, discards irrelevant molecules such as buffers and solvents, and documents every source in a report. An interactive viewer then lets scientists plot potency against rotatable bonds, showing that the most rigid compounds tend to be the most potent.
From there the agent runs matched molecular pair analysis, comparing molecules that differ only by cyclisation, and trains potency models across four experimental readouts, validating them on held-out data. Finally it proposes new rigidified macrocycles, explains its medicinal-chemistry rationale, predicts each design’s potency, and suggests synthesis routes consistent with documented procedures.
Access is straightforward: individual users download the Antigravity app and install the Science plugin, while enterprise users sign in with Google Cloud credentials through a business account. Applications for the free six-month top tier are open now, with scientists onboarded on a rolling basis subject to eligibility, capacity and selection.
Why It Matters
Drug discovery is slow largely because of data wrangling, not because scientists lack ideas: curating structures from papers, figures and catalogs, then training models, can consume days per iteration. Compressing that loop to about an hour lets researchers test more hypotheses per week, which matters most in fields like oncology, where the target lung cancer spreads to the brain.
The release also extends Antigravity beyond software engineering, where Google has been consolidating its coding tools, into laboratory science. That widening matters because agentic platforms are competing on real workflows, not demos: the platform that handles a chemist’s literature, data and modelling in one place has a stronger claim on research budgets than one that only writes code.
Our read
Our read: DeepMind is careful to say these designs are hypotheses, not drugs, and that caveat is the important line in the announcement. Sparse data, assay noise between labs and narrow measurement ranges cap predictive precision, so proposed macrocycles must be synthesised and tested before they mean anything clinically — a standard AI drug-discovery efforts do not always meet.
The free tier for 10,000 scientists is the commercial tell: DeepMind wants working chemists, not just press coverage, inside Antigravity before rivals lock up laboratory workflows. If those scientists stay after six months, Google gains a foothold in research software; if the hour-long workflow proves fragile on messier real data, the offer will read as marketing.
Background
Antigravity is Google DeepMind’s agentic development platform, previously extended with a plugin marketplace as Google shut down its Jules coding agent in October. The cheminformatics release follows an earlier Antigravity for Science demonstration that searched for antimicrobial peptides, moving the platform from code and text into experimental science.
Source: DeepMind’s announcement
0 Comments