DARPA BTO director: virtual cell models and de novo protein design don't work
DARPA Biological Technologies Office director Mike Koeris publicly confirmed his assessment that virtual cell models and de novo protein design do not work yet, while predicting protein sequencing will be solved within a decade.
Mike Koeris has headed DARPA's Biological Technologies Office (BTO) since April 2024. BTO allocates a share of the U.S. defense agency's multibillion-dollar biotech budget. Before DARPA, Koeris trained in the lab of Jim Collins, one of the founders of synthetic biology (Collins built the first genetic toggle switch in E. coli in 2000), and founded two biotech startups, both of which he later sold. At SynBioBeta 2026 panels in May he publicly criticized two of the most heavily promoted AI directions in biology.
Virtual cell models don't work right now. His exact words. A virtual cell is a computational model meant to predict how a cell responds to a drug without a lab experiment. The approach is championed by Chan Zuckerberg Initiative (Mark Zuckerberg's organization) and the biomedical Arc Institute, with support from Nvidia. Whole-cell modeling is officially listed among BTO's own goals, so Koeris is criticizing a direction his own office funds.
De novo protein design doesn't work either. This means creating protein structures with a specified function that do not exist in nature. David Baker received half the 2024 Nobel Prize in Chemistry for a method of such design. The same generative approach recently failed against tau fibrils in Alzheimer's disease, although it had worked on other targets. According to Koeris, current algorithms do not build protein structure from true scratch: they take an existing natural scaffold, adjust it slightly, and call that design.
Koeris assessed a third challenge from the same field differently. He predicts that the problem of protein sequencing (deciphering an unknown amino acid sequence without a reference for comparison) will be fully solved within ten years and will cost as little as DNA sequencing. Behind this prediction sits a concrete program: BTO launched PROSE in 2026 with a measurable target of reading proteins of 300 amino acids or longer at above 99% per-residue accuracy. The distinction from protein design is fundamental: sequencing reads a sequence already present in nature, while design must produce a structure that has never existed.
Koeris stated his risk philosophy in 2024: roughly half of the projects DARPA funds may fail, but that yields enormous learning, and the other half produces results on the scale of GPS, the internet, and mRNA vaccines. DARPA landed in the successful half in 2011 when it funded early mRNA vaccine technology through the ADEPT program; Moderna, then a little-known startup, took part in that work.
On September 28, SynBioBeta founder John Cumbers published exact quotes from the May panels. The next day, Koeris responded on LinkedIn: "I stand by these quotes. Onward. We do this because it's hard. We do this for the mission!" This is the assessment of someone who controls part of DARPA's budget and decides which programs receive funding.
Open the Chan Zuckerberg Initiative page on Eternal Search
Sources
[1] x.com
[2] nobelprize.org
[3] t.me
The news item explicitly names Chan Zuckerberg Initiative as one of the main advocates of virtual cell models, which is exactly the technology Koeris says does not yet work. The Eternal Search organization page connects the policy signal to the specific actor promoting the approach, giving readers a concrete place to examine who is behind this direction rather than leaving the name as a floating reference.