Microsoft's Quine turns a pancreatic cancer question into a weekend of lab picks

Microsoft Research has introduced Quine, an experimental AI system for biology that couples one model trained on many kinds of biological data with a tool that proposes experiments and learns from the results. The company says it is for research only, and that access is limited to a fellows program and a few collaborations.

One model, many kinds of data

According to Microsoft, Quine has two parts. The first is a world model of biology, trained jointly on genomics, proteins, chemistry, cellular state and bioimaging data. The idea is that evidence from one kind of data can inform predictions about another.

The second is an interactive harness that connects that model to reasoning, scientific tools and the literature. Scientists pose a question, and Quine returns ranked proposals for experiments to try.

A loop with the lab

The wet-lab results then go back into the system. Microsoft describes a continuous loop between computation and experiment, which refines both the models and the next questions.

In its main example, Microsoft says Quine predicted and prioritized compounds that could shift the state of pancreatic cancer cells. The company says the process "took just one weekend, potentially saving months" of work. It also says the experiments showed unexpected compound effects, suggesting a richer cellular landscape than was previously understood.

The company also claims that training across modalities strengthens performance compared with orchestrating separate single-domain models. The post, as summarized here, gives no detailed benchmarks for that claim.

Guardrails and who gets it

Microsoft calls Quine "experimental research technology intended only for research, not clinical or medical use." It adds that outputs may be incomplete or inaccurate and need review by qualified researchers.

For now, access runs through the Quine Fellows program for qualified scientists. Microsoft says it plans to expand availability through its Microsoft Discovery product.

The pancreatic cancer result is Microsoft's own account of its own tool. What would settle the matter is outside groups reproducing the compound predictions, and a published comparison against simpler pipelines.