Evogene ChemPass explores <0.1% chemical space

Evogene is showing up in AI-biotech conversations because its ChemPass AI platform sits at a specific point in the drug-discovery workflow: not at the clinical stage, but at the earlier step where researchers try to find and optimize small molecules worth testing. The company says the platform is designed to search very large chemical space, generate new candidate structures, and optimize multiple properties at once rather than one after another. Social posts on Wednesday framed that as part of a broader industry argument over whether AI should start from known safe chemistry or generate new molecules from scratch, but Evogene’s own materials describe a hybrid process that uses both purchased compounds and de novo design. ### Why are people talking about “chemical space” in the first place? Drug discovery uses “chemical space” as shorthand for the huge universe of possible small molecules. Review literature describes that universe as vastly larger than any library that can be synthesized or tested in a lab, with estimates above 10^60 for drug-like molecules. That gap is the reason companies use computational methods to narrow the search before doing experiments. (evogene.com) Evogene’s March 2026 BIO-Europe Spring materials said ChemPass AI is designed to “explore vast chemical space” and generate novel, potent molecules optimized across multiple parameters. The company’s website says those parameters include potency, selectivity, stability, toxicity, ADME and synthesizability. ### So what exactly does ChemPass do? (nature.com) Evogene describes ChemPass AI as a closed-loop discovery engine. The process starts with literature search, data preprocessing and target modeling, then moves into virtual screening of commercially available compounds, followed by assay testing of the best performers. After that, the platform generates new molecules optimized across multiple criteria, and those novel candidates are synthesized and tested. (evogene.com) That matters because the company is not presenting ChemPass as a pure text-to-molecule generator operating in isolation. Its own workflow combines computational screening, iterative lab validation and then de novo molecular design. That is narrower, and more concrete, than the broader social-media framing that treats the field as a simple choice between repurposing known compounds and generating entirely new ones. That characterization is an inference from Evogene’s published workflow. (evogene.com) ### What numbers has Evogene put behind the platform? Evogene said on June 10, 2025 that its foundation model for small-molecule design was built on a dataset of about 38 billion molecular structures and now serves as a core component of ChemPass AI. In the same release, the company said internal computational analysis showed about 90% precision in “successful and precise, novel molecule designs,” compared with about 29% for “traditional GPT AI-model.” (evogene.com) Those figures are company-reported and described as internal computational analysis, not clinical or regulatory validation. Evogene’s current website repeats the 38 billion-structure universe and the roughly 90% precision figure, again in the context of platform performance rather than approved medicines. ### Where does Google Cloud fit into this? Evogene and Google Cloud expanded their collaboration on February 10, 2026 to integrate AI agents into ChemPass AI using Vertex AI. (evogene.com) Evogene said the goal was to automate and scale complex discovery workflows, enable parallel molecular exploration, and shorten design-make-test-analyze cycles. (evogene.com) Evogene said on May 20, 2026 that the February Google Cloud work was part of its effort to keep advancing ChemPass AI after a strategic transformation launched in 2025. CEO Ofer Haviv said the company is focused on execution, tech-engine development and expanding its pharma and agriculture pipeline through collaborations. ### Has this produced outside partnerships, or is it still mostly a platform story? (evogene.com) Evogene said in its first-quarter 2026 results that it added three pharma collaborations in the quarter: with Systasy Biosciences together with LMU University Hospital Munich, with Queensland University of Technology, and with Unravel Biosciences. The company said those additions brought its total number of publicly disclosed collaborations in the pharma domain to four. (evogene.com) The next public datapoints are likely to come from those collaborations or from future company updates rather than from Wednesday’s social-media discussion. Evogene’s May 20, 2026 results release and its February and March platform announcements remain the clearest primary-source record of what ChemPass AI is, what numbers the company attaches to it, and where it says the program is headed. (evogene.com)