Microsoft Introduces Quine, an AI Biology System Tested on Tumor-Cell Shifts

Microsoft Research introduced Quine on September 29 as an AI system meant to help scientists decide which biological experiments to run. In a collaboration with Broad Institute researchers, it used the system to rank compounds for tests on pancreatic cancer cells. Microsoft says the top-ranked compounds produced the intended shifts in lab assays, but Quine remains an experimental research tool, not a clinical one.

A search problem in the lab

The cancer study began with a question about cell state: whether features of a tumor beyond its genetics could be useful targets for research. Working on pancreatic ductal adenocarcinoma, Microsoft and Broad researchers used Quine to predict how thousands of compounds might shift tumor cells between states. Those predictions gave the team a way to choose a smaller set for physical testing.

Quine is designed for more than a single kind of biological data. Its world model learns from information including genetic sequences, protein structures, cell states and images. Microsoft pairs that model with a system connecting reasoning tools, scientific literature and researchers. The goal is to use evidence at one biological scale to inform predictions at another, then bring experimental results back into the research process.

The weekend search and the assay results

Microsoft says Quine narrowed the compound search and prioritized a handful of candidates for lab validation in one weekend. That is the time it reports for selecting what to test, not for completing the experiments or developing a therapy. Some compounds with strong effects worked through mechanisms the team had not expected, giving researchers further leads to investigate.

In assays focused on moving cells from a classical state toward a basal state, the highest-ranked compounds produced the largest intended shifts, according to Microsoft. Movement in the reverse direction was weaker, as Quine had predicted. The experiments also found cells moving toward a distinct third state after treatment with several compounds—another result the system had predicted. That observation suggests the researchers’ initial two-state picture did not capture the full range of responses.

Access opens; the next experiments remain

Applications for the first Quine Fellows cohort are open, but initial access is limited to that program and select research collaborations. Microsoft says it expects to broaden access through products such as Microsoft Discovery as the technology matures. For now, Quine is a research system whose outputs may be incomplete or inaccurate. Qualified researchers must review its suggestions and validate them through appropriate scientific and experimental work.

The unexpected third-state response gives the team a concrete next question. Microsoft says it plans to incorporate new RNA data and tasks to represent that more complex landscape and improve predictions about state changes. It also aims to provide confidence estimates that could help scientists choose which hypotheses deserve lab time. The assays show Quine helped guide this search; they do not make the tested compounds treatments.