Google DeepMind released Cheminformatics Skills for the Google Antigravity platform

Google DeepMind has released a set of tools called Cheminformatics Skills for the Google Antigravity platform, intended for both computational and laboratory chemists. According to the company, the goal is to eliminate time spent manipulating cheminformatics data - in practice, the same molecule appears in many forms (an IUPAC name in an article, a Markush structure in a patent, an SDF file in a catalog), and a conversion error can end up describing a completely different compound.

The architecture separates the role of the language model from the computational tools: the model handles workflow orchestration, database navigation, and scientific reasoning, while the actual molecular computations are carried out by deterministic tools. The Skills are built on the open-source RDKit toolkit and connected to the ChEMBL and PubChem databases; according to the company, established best practices from the field are also encoded into the agentic workflow.

As an example, the company describes a technique called molecule rigidification in medicinal chemistry - adding a bond that restricts the molecule's flexibility and thereby increases its potency when binding to a target. A concrete example given is the transition from the drug crizotinib (five main rotatable bonds) to the drug lorlatinib, which was cyclized so that the number of rotatable bonds dropped to zero, leading to higher potency.

The source covers only this one technical procedure as an illustration of the agents' capabilities; further details on availability, pricing, or other features of Cheminformatics Skills are missing from the source.