Roughly 37,000 AI agents, organized like a biopharma company, proposed a therapeutic strategy that humans still need to test.
That distinction matters more than it might seem. The system’s proposed target, CD276, later attracted FDA breakthrough therapy designation when Merck and Daiichi Sankyo pursued a nearly identical AI-powered websites antibody-drug conjugate approach for extensive-stage small-cell lung cancer, which is the closest independent validation a computational hypothesis can receive without entering a lab.
What the Virtual Biotech Actually Did
The system mirrors a corporate org chart built entirely from software, with thousands of specialized agents coordinated by “chief scientist” AI at the top.
Think of the architecture as a corporate org chart built entirely from software. Low-level agents extracted endpoints from clinical trial reports and normalized genomic data.
Mid-level agents synthesized those outputs into integrated target profiles. At the top, “chief scientist” coordinators decided which hypotheses to pursue and commissioned new analyses from below.
That structure processed 55,984 clinical trials, turning heterogeneous reports into structured, searchable evidence. The system then integrated statistical genetics, single-cell data, spatial omics, and clinicogenomic datasets. That multimodal evidence was used to evaluate CD276, also known as B7-H3, as a lung-cancer target.
According to the Virtual Biotech preprint published in September 2026, the agents “evaluated B7-H3 as a lung cancer target, integrating statistical genetics, single-cell, spatial, and clinicogenomic evidence to propose an antibody-drug conjugate strategy.”
The meta-analysis behind that proposal carries real weight. Drugs targeting cell-type-specific genes, the kind CD276 represents in tumor cells, are 40% more likely to advance from Phase I to Phase II trials, 48% more likely to reach market, and show 32% lower adverse-event rates compared with non-cell-type-specific targets, according to the same preprint.
Discovery or Strategy? Why the Distinction Matters
The AI proposed where to aim; it did not design, synthesize, or clinically test any new molecule.
The AI did not synthesize a new molecule. It did not run experiments in a lab or initiate a clinical trial.
Lung Cancer Europe confirmed what the preprint itself makes clear: the system did not discover CD276, and no treatment has been tested in humans as a direct result of this work. The authors describe their contribution plainly: “human-guided multi-agent systems can conduct transparent, multiscale analyses to inform therapeutic-development decisions.”
That framing places the Virtual Biotech in a specific lane of AI drug discovery, one focused on evidence integration and strategy recommendation rather than molecular design. Contrast that with Insilico Medicine’s rentosertib, an AI-designed molecule targeting a novel protein called TNIK for idiopathic pulmonary fibrosis.
That program went from AI-identified target to Phase IIa trial results. In Nature Medicine in June 2025, 71 patients showed a 98.4 mL improvement in lung function.
By mid-2026, Insilico had initiated a Phase III trial. That is what an AI-designed drug advancing through clinical validation looks like.
The Virtual Biotech represents a different but complementary capability: a system that reads the entire clinical-trial landscape and tells you where to aim, not what molecule to fire. The convergence with Merck and Daiichi Sankyo’s CD276 ADC program, which received FDA breakthrough therapy designation and priority review, suggests the AI’s reasoning was sound.
Whether pharma companies will begin trusting algorithmic chief scientists to set research agendas at scale is the question the next generation of these systems will have to answer.
