Google Antigravity applies agentic tools to cheminformatics

DeepMind presented a workflow in which Google Antigravity, equipped with new Cheminformatics Skills, supports several stages of early drug discovery. In a case involving treatments for lung cancer, the agent searched scientific literature and other sources, extracted and curated thousands of compounds, removed irrelevant substances, and recorded their origins. The workflow was completed in about an hour, compared with the multiple days often required for one iteration.

The system combines a language model for workflow coordination, database navigation and scientific reasoning with deterministic chemistry software. Built around RDKit and connections to ChEMBL and PubChem, it can display compound data, examine links between potency and molecular flexibility, perform matched molecular pair analysis, and train predictive models. The analysis found that cyclization generally reduced rotatable bonds and increased potency. The example compared crizotinib, which has five major rotatable bonds, with the more rigidified macrocycle lorlatinib.

The agent also proposed new macrocycles and possible synthesis routes, while generating reaction diagrams and potency predictions. DeepMind stressed that these designs are hypotheses requiring synthesis and experimental validation because sparse data, assay variation and limited measurement ranges constrain model accuracy. The skills are available in Google Antigravity, and applications are open for six months of free access at the highest usage tier for 10,000 scientists, subject to eligibility and capacity.