Microsoft Quine Uses Multimodal AI to Accelerate Biology and Cancer Research
Microsoft's Quine links a multimodal biology model to research tools, prioritizing pancreatic cancer compounds in one weekend and validating them in labs.
Summary
Microsoft Research introduced Quine on September 29, 2026, as an experimental system combining a multimodal biology world model with an interactive harness linking reasoning and orchestration models, scientific tools, literature, wet labs and researchers. It jointly learns representations spanning genomics, proteins, chemistry, RNA, cellular state, structure, function and bioimaging, aiming to predict how biological systems respond to interventions and prioritize experiments across scales rather than replace laboratory validation.
Working with the Broad Institute of MIT and Harvard, Microsoft used Quine to rank thousands of compounds for shifting transcriptional states in pancreatic ductal adenocarcinoma, the most common pancreatic cancer. The highest-ranked candidates produced the largest intended classical-to-basal shifts across multiple wet-lab assays. Narrowing the search to several candidates took one weekend, potentially avoiding months of work and significant costs. Compounds with unexpected mechanisms showed potential for repurposing and discovery. Quine correctly predicted weaker basal-to-classical effects and identified movement toward a third phenotype, which experiments confirmed, challenging the simple classical-basal model.
Microsoft will add RNA datasets and tasks, strengthen transition predictions and provide calibrated confidence estimates. Applications are open for the first Quine Fellows cohort, with initial access limited to fellows and selected collaborations under safeguards and internal review. Quine is for research only, not clinical or medical use, and may produce incomplete or inaccurate outputs requiring expert review and experimental validation. Microsoft expects eventual access through products such as Microsoft Discovery.
Positives
- One weekend was sufficient for Quine to narrow thousands of pancreatic cancer compounds to several candidates for wet-lab validation.
- Quine's highest-ranked compounds produced the largest intended classical-to-basal tumor-state shifts across multiple assays.
- Wet-lab experiments confirmed Quine's prediction that several compounds would move cells toward a distinct third phenotype.
- Unexpected compound mechanisms provided early evidence of opportunities for drug repurposing and discovery.
- The Quine Fellows program opens the experimental system to scientists who can accelerate research and guide its development.
Risks & concerns
- Quine may generate incomplete or inaccurate outputs, requiring qualified researchers and appropriate experimental validation.
- Basal-to-classical tumor-state transitions remained difficult, with available compounds producing weaker effects.
- Initial access is restricted to Quine Fellows and selected research collaborations.
- Quine remains experimental and is not approved or intended for clinical or medical use.
- The newly observed third phenotype shows that pancreatic cancer states are more complex than a classical-basal axis.