Google Antigravity cuts drug discovery loop to about an hour

The announcement builds on a long-standing bottleneck in medicinal chemistry: molecules appear in many formats — an IUPAC name in a paper, a Markush structure in a patent, an SDF in a catalog — and a single altered atom or bond describes a different compound entirely. DeepMind's Cheminformatics Skills are designed to unify those sources while keeping deterministic tools like RDKit in charge of molecular calculations, with the language model orchestrating the workflow and reasoning about the science.

The walkthrough centers on rigidification, a classic tactic to improve potency by tying down a floppy molecule's rotatable bonds. DeepMind uses crizotinib, a first-generation tyrosine kinase inhibitor for non-small cell lung carcinoma with five major rotatable bonds, and lorlatinib, a third-generation analogue cyclized to zero rotatable bonds, as the reference case. From there the agent extracted compounds from literature, flagged that the macrocycles are largely predicted to be brain-penetrant — relevant because the target lung cancer often metastasizes to the brain — and built potency models validated on held-out data.

DeepMind is explicit about limits: sparse data, inter-lab assay noise and narrow dynamic ranges constrain even the best models, and designs must be synthesized and validated experimentally. The commercial signal is in distribution: the skills ship today inside Antigravity, enterprise users can log in with Google Cloud credentials, and a free six-month top-tier offer for 10,000 scientists is meant to pull researchers onto the platform.