AI platform screened 69 billion compounds

Bret Kerr said in a September 2 post on X that an AI platform screened 69 billion compounds to identify cancer drug leads, naming NVIDIA’s BioNeMo and describing an agent-style workflow of hypothesis generation, simulation and validation. The post did not identify a paper, company lab, university group or cancer program behind the claimed screen. A review of NVIDIA materials and recent scientific literature shows BioNeMo is positioned for drug-discovery workflows, but the specific 69 billion figure is already associated with a separate virtual-screening stack built around Enamine REAL Space. ### Where does the 69 billion figure come from? Nature Biotechnology published a paper on September 1 describing AdaptiveFlow, an “AI-enhanced adaptive virtual screening” platform that provides access to Enamine REAL Space, which the paper says comprises 69 billion compounds. The paper says the platform was used to explore that library and identify inhibitors of ferroptosis suppressor protein 1, or FSP1. PubMed Central’s copy of the preprint and earlier VirtualFlow materials also tie the 69 billion number to Enamine’s ready-to-dock library, not to a standalone NVIDIA announcement. (nvidianews.nvidia.com) The published record therefore supports one narrow point: 69 billion is a real number in current virtual-screening literature. It does not, on the sources reviewed, establish that NVIDIA itself announced screening 69 billion compounds for cancer leads on September 2. ### What has NVIDIA actually said about BioNeMo? NVIDIA said on June 23 that it launched the BioNeMo Agent Toolkit, which the company described as a set of tools for agentic workflows across biology, chemistry, genomics and drug discovery. (nature.com) NVIDIA said the toolkit lets agents gather evidence, reason across findings, run computational experiments and recommend next steps, and listed tasks including protein structure prediction, molecular docking, generative chemistry and biomarker discovery. NVIDIA’s BioNeMo product pages separately describe the platform as supporting target identification, lead optimization, virtual screening, docking and property prediction. A January 12 NVIDIA announcement said more than 50 companies were using the toolkit and that BioNeMo had been expanded for “lab-in-the-loop” drug-discovery workflows. Neither of those company statements, in the material reviewed, mentions a 69 billion-compound cancer screen. (nvidianews.nvidia.com) ### Does “screened 69 billion compounds” mean 69 billion molecules were tested in a lab? Virtual screening refers to computationally searching large chemical libraries for molecules predicted to bind a target, not physically testing every compound in a wet lab. Nature’s virtual-screening overview describes it as a computational approach to identify structures likely to have desired properties, and recent reviews say ultra-large virtual screening can evaluate hundreds of millions to billions of compounds before a much smaller set is synthesized or experimentally tested. (nvidianews.nvidia.com) The AdaptiveFlow paper follows that pattern. The authors describe a screening-ready library of 69 billion compounds and report experimentally validated FSP1 inhibitors after narrowing candidates through the computational workflow. That is different from saying 69 billion compounds became drug leads or entered animal or human testing. ### So what can be stated with confidence right now? (nature.com) The verifiable facts are that Bret Kerr posted the claim on September 2, NVIDIA markets BioNeMo for agentic drug-discovery workflows, and peer-reviewed literature published September 1 links a 69 billion-compound screen to AdaptiveFlow and Enamine REAL Space. The unverified part is the implied connection that a specific BioNeMo-driven cancer-lead campaign screened all 69 billion compounds. (nature.com) September 1 is the next concrete milestone in the record because that is when Nature Biotechnology published the AdaptiveFlow paper, and June 23 is the date of NVIDIA’s BioNeMo Agent Toolkit announcement. Any stronger claim tying the 69 billion screen directly to BioNeMo would require a paper, conference talk, institutional release or named research team that has not surfaced in the sources reviewed. (nature.com) (nvidianews.nvidia.com)