Microsoft Research Debuts Quine: A Multimodal World Model of Biology
Microsoft Research has introduced Quine, a multimodal world model of biology that aims to represent and predict the state of biological systems. The system is designed to work in conjunction with researchers and experimentalists to accelerate the discovery process.
Quine's core model is trained on various data types, including sequence, structure, function, cellular state, and imaging, allowing it to learn representations shared across biological scales. This approach enables joint training across different modalities, which strengthens rather than dilutes performance.
The system was tested in a specific case: pancreatic ductal adenocarcinoma (PDAC), the most common form of pancreatic cancer and one of the hardest to treat. Researchers used Quine to rank compounds for their potential to move tumor cells between states relevant to therapy, with several top-ranked candidates validated across multiple wet-lab assays.
The results showed that Quine's predictions matched experimental outcomes, with compounds ranked highest moving cells furthest in the intended direction. This indicates that AI can surface opportunities for drug repurposing and discovery, potentially saving months of experimental work and considerable research cost.