Microsoft researcher says AI is learning the language of biology

Ava Amini, a principal researcher at Microsoft Research, has been making the case for artificial intelligence as a powerful new tool for biology, telling a packed “Lectures on Tap” event in Cambridge, Massachusetts, that computation is helping scientists make sense of biological complexity at a scale humans cannot manage alone.

Amini said AI is increasingly able to spot patterns in vast biological datasets that would otherwise be impossible to interpret. “Computation gives us this incredibly powerful toolkit to understand what I think is the most complex and intricate system that we have, which is the system and the language of biology,” she said. By way of example, she noted that a single cancer biopsy can generate close to 50 million individual data points.

Microsoft said it is developing a range of generative models aimed at working directly in what researchers describe as the “language” of biology. These include EvoDiff and The Dayhoff Atlas, which are designed to create new proteins by prompting models in biological sequences rather than traditional code. Amini said the Dayhoff models lifted the success rate for producing new proteins to about 50%, up from 16% using earlier approaches.

Crucially, she said the work is not limited to simulations. Microsoft has tested AI designed proteins in the laboratory and observed that they perform the intended functions. “We’ve actually gone and measured and tested in the lab in the real world to show that these proteins have the functions that we meant and sought to have,” Amini said.

The company also pointed to limits in current approaches, noting that many AI models of whole cells tend to predict average values and stop improving as more data is added. Recent studies, including work led by Amini’s team, have highlighted these constraints.

Amini co-leads Project Ex Vivo, a research collaboration between Microsoft and the Broad Institute, supported by the Dana-Farber Cancer Institute. The project is focused on building a precision oncology framework that tightly links experimentation with computation. “As a technologist, we use these findings as fuel, and we want to take as much as we can to actually go further,” Amini said.

The Recap

• Microsoft researcher outlines AI learning biology's language to reimagine medicine.
• Dayhoff models raised protein design success from 16% to 50%.
• Project Ex Vivo builds precision oncology framework integrating computation and experiments.