John Jumper

For 60 years, scientists had puzzled over the protein folding problem: There is a vast number of ways a protein can fold, making it difficult to accurately predict. But understanding the relationship between a protein’s 1-dimensional code and its 3-dimensional structure is key to learning more about diseases like Parkinson’s and designing drugs to combat viruses like HIV.

Then, in late 2020, a team from Google DeepMind, led by John Jumper, then a senior research scientist at the organization, cracked it using a machine learning algorithm they called AlphaFold 2. Jumper’s team later released the model on the internet for free.

“What I'm most proud of is the extent to which it's making all of structural biology five or ten per cent faster,” Jumper says. Previously, biologists had to spend years meticulously observing a protein and documenting its structure before starting their experiments, but “Al​​phaFold, in many cases, lets them skip that one or two years,” helping accelerate research on a range of issues from better understanding our cardiovascular system to tackling antibiotic resistance.

In May 2024, Jumper’s team unveiled AlphaFold 3, which in addition to proteins, can predict other molecules like DNA and RNA. Unlike its predecessor, AlphaFold 3 was not made open source. “There were certainly commercial considerations involved in this,” Jumper says. “At the same time, we’ve made a commitment to make the model available for non-commercial use. I think we’ve struck quite a reasonable balance.”