Microsoft Research Introduces Quine, a Research-Only AI System for Studying Biology

What is Quine?

According to Microsoft Research, Quine is a research program with two goals. The first is to build a multimodal "world model" of biology, meaning one that learns from many kinds of biological data at once. The second is to pair it with an interactive "harness": software that links the model to scientific tools, published literature and the researchers using it. The announcement was written by Nicolo Fusi, VP and Distinguished Scientist, and Jonathan M. Carlson, Vice President.

By world model, Microsoft Research means a system that can do three things. It can represent the current state of a biological system, predict how that state will change after an intervention such as adding a drug, and reason about the consequences several steps ahead. Microsoft says plainly that such systems will not replace experiments. Instead, they use computation to explore, propose and rank possible directions before scarce lab resources are committed. Microsoft also says a world model will never perfectly model biology; it only needs to usefully inform how experiments are designed.

Read the full description

Sources are collected, independently checked, then reviewed by Jev.

Two core components

The pancreatic cancer case: what results did Microsoft report?

Microsoft Research used pancreatic ductal adenocarcinoma (PDAC) as its example. Microsoft describes PDAC as the most common form of pancreatic cancer and one of the most challenging to treat. PDAC tumor cells can be in different "states", two of which are called classical and basal. Microsoft says these states are associated with how the cells respond to treatment. Working with Broad Institute researchers, Microsoft used Quine to predict and rank thousands of compounds by their potential to shift tumor cells between these states.

The predictions were then checked in wet-lab experiments. These are tests on real cells in a laboratory, as opposed to computer simulations, and they used several different lab tests, known as assays.

Microsoft says the whole process took just one weekend, from narrowing the compound search space to picking a handful of candidates for lab validation. It says this could potentially save months of experimental work and significant research costs. Microsoft also notes that some of the strongest effects came from compounds with unexpected mechanisms of action. It sees this as early evidence that AI could help with drug repurposing (finding new uses for existing compounds) and with discovering new drugs.

Who can use it? Access and future plans

According to Microsoft Research, access to Quine is being opened in careful, phased steps. At first it is limited to the Quine Fellows program and select research collaborations. Microsoft says oversight will come from ongoing internal review and safeguards built into the system. Applications for the first group of Quine Fellows are open now. Microsoft says it expects to broaden access through products such as Microsoft Discovery as the technology matures.

Microsoft also says Quine has been used in its own research on cancer biology, protein engineering, genomics and bioimaging. Its next steps are to add new RNA datasets, strengthen its predictions of how cells change state, and provide calibrated confidence estimates. These are scores showing how much trust each prediction deserves, so scientists can decide which ideas to test first.

What it means for general readers

• No direct effect on doctor visits or medication in the near term: Microsoft states explicitly that Quine is not for clinical or medical use.
• The change worth watching is in how research is done. If Microsoft's account holds, AI could first help scientists screen large numbers of candidate compounds, so labs can spend their time on the most promising directions.
• Treat the results with caution: they currently come from a single source, Microsoft, and still await more independent research and peer review.

Browse the latest news in this topic

• AI news