Microsoft Research Introduces Quine, an AI Research System Designed for Biological Complexity

The system integrates a multimodal world model that learns shared representations across diverse biological scales and modalities, including sequence, structure, function, cellular state, and imaging data. This design allows the model to use evidence from one modality to inform predictions in another, capturing complex biological relationships that specialized single-domain models might miss. Quine connects this model to a harness of scientific tools, literature, and researchers to support an iterative scientific process.

In a practical application involving pancreatic ductal adenocarcinoma (PDAC), researchers used Quine to predict and prioritize compounds capable of shifting tumor cells between therapeutically relevant states. The process narrowed a massive search space to a handful of promising candidates for wet-lab validation in just one weekend. The experiments not only confirmed the model's hypotheses regarding cell-state transitions but also generated new insights, such as identifying a third phenotype in the pancreatic cancer cell-state landscape.