Google Antigravity Turns Days of Drug Discovery Into One Hour
Google Antigravity’s new chemistry agent compresses days of lung cancer drug analysis into one hour, but every proposed molecule still needs lab validation.
Summary
On November 3, 2026, Google made Cheminformatics Skills available in Google Antigravity through the Science plugin, including to Enterprise users signing in with Google Cloud credentials. The agent compresses a drug discovery cycle that usually takes multiple days into about one hour: it mines literature and vendor records, retrieves thousands of compounds, removes buffers and solvents, links structures to assays and sources, and opens an interactive viewer with metadata and scatter plots. The language model orchestrates reasoning and database work, while deterministic RDKit tools calculate chemistry through ChEMBL and PubChem connectors, unifying IUPAC names, Markush structures and SDF files where one changed atom or bond can define another compound.
The lung cancer demonstration rigidified crizotinib, a first-generation small-molecule TKI for some non-small cell lung carcinomas with five major rotatable bonds, toward lorlatinib’s third-generation macrocyclic architecture with effectively zero. The curated macrocycles were largely predicted to cross the blood-brain barrier, relevant because this cancer often spreads to the brain; plots associated greater rigidity with higher cellular potency. Matched molecular pair analysis found cyclization generally cut rotatable bonds and raised potency. The agent trained many potency models across four experimental readouts, validating held-out data and favoring chronological splits when dates existed or scaffold splits for generalizability and out-of-distribution testing. It then proposed novel macrocycles, reaction mechanisms with curved-arrow diagrams, documented synthesis routes and predicted potencies, with outputs exportable to standard chemistry software and candidates ready for AlphaFold or molecular dynamics analysis.
Sparse data, inter-lab assay noise and narrow dynamic ranges still limit predictive precision, so every computational design requires synthesis and experimental validation. Applications are open for 10,000 eligible scientists to receive Antigravity’s highest usage tier free for six months, with rolling onboarding subject to capacity and selection.
Positives
- About one hour replaced a workflow that previously required multiple days and substantial computational expertise.
- Thousands of compounds were retrieved, filtered, sourced and organized within minutes for interactive exploration.
- Five major rotatable bonds in crizotinib fell to effectively zero in lorlatinib’s more potent macrocyclic architecture.
- Four experimental readouts supported potency modeling with held-out validation and chronological or scaffold splits.
- 10,000 eligible scientists can receive the highest Google Antigravity usage tier free for six months.
Risks & concerns
- Sparse chemical data restricts the predictive precision of even powerful molecular models.
- Inter-lab assay noise can obscure whether predicted potency improvements will reproduce experimentally.
- Narrow dynamic ranges limit model discrimination among candidate molecules.
- A single incorrect atom or bond can turn a promising compound into a chemically different, inactive structure.
- Every proposed macrocycle remains a hypothesis requiring synthesis and laboratory validation.
