AI System Identifies Lung Cancer Target with 40% Higher Success Rate
A team of researchers has developed an AI system consisting of 37,000 agents that analyzed 55,984 clinical trials to identify CD276 as a promising target for lung cancer treatment. The system's proposed target was later validated when Merck and Daiichi Sankyo received FDA breakthrough therapy designation for their nearly identical antibody-drug conjugate approach.
The AI system, called Virtual Biotech, mirrors the structure of a corporate org chart built entirely from software. It consists of low-level agents that extracted endpoints from clinical trial reports and normalized genomic data, mid-level agents that synthesized these outputs into integrated target profiles, and high-level 'chief scientist' coordinators that decided which hypotheses to pursue.
The system processed vast amounts of evidence from various sources, including statistical genetics, single-cell data, spatial omics, and clinicogenomic datasets. This analysis revealed that drugs targeting cell-type-specific genes, such as CD276, are 40% more likely to advance from Phase I to Phase II trials and have 32% lower adverse-event rates compared to non-cell-type-specific targets.
The Virtual Biotech system's contribution lies in its ability to conduct transparent, multiscale analyses and inform therapeutic-development decisions. It did not design or synthesize new molecules but rather identified where to aim based on the evidence it analyzed.