Quine

Microsoft Research's AI research system for biology, announced by Nicolo Fusi and Jonathan M. Carlson and filed on the blog under Microsoft Health Futures. It pairs a multimodal world model of biology with an interactive harness that connects the model to orchestration and reasoning models, scientific tools, the literature, the wet lab and the scientists using it. Microsoft defines the world model as a system that represents the state of a biological system, predicts how that state changes under interventions, and reasons over the consequences several steps ahead. It learns shared representations across sequence, structure, function, cellular state and imaging data, trained jointly on genomics, proteins, chemistry, RNA and cell state, and bioimaging rather than by orchestrating separate single-domain models, and Microsoft says the joint training strengthens rather than dilutes performance. The project page traces Quine to six lines of earlier Microsoft work: proteins, the regulatory genome, the tumor microenvironment, tissue pathology, microscopy and biological data.

The worked example is pancreatic ductal adenocarcinoma, studied with the Broad Institute of MIT and Harvard to test whether a tumor cell's transcriptional state can be a drug target. Quine predicted and ranked thousands of compounds by their potential to shift tumor cells between states. In wet-lab assays of the classical-to-basal transition, its top-ranked compounds produced the largest intended shifts, and the process from narrowing the compound space to picking candidates for validation took one weekend. It also predicted that the reverse, basal-to-classical shift would be weaker, and that several compounds would push cells toward a distinct third phenotype; the experiments bore out both. Some of the strongest effects came from compounds with unexpected mechanisms of action.

This is an announcement only: Microsoft published no technical report, model size, benchmark results or weights. Access is limited to select research collaborations and the Quine Fellows program, a 16-week fellowship hosted by Microsoft Research in Cambridge, Massachusetts (applications September 29 to November 2, 2026; fellowship June 7 to September 24, 2027) for PhD candidates, postdocs and academic or independent researchers, with interests including protein and enzyme design, cell-state perturbation and early-stage therapeutics for under-resourced diseases. Microsoft calls Quine experimental and not for clinical use, and expects to widen access through products such as Microsoft Discovery.