Why is it so hard to find the shape of a protein?
Proteins are the workers of the body: some digest food, some carry oxygen, some defend us against germs. Each one is a long chain of small building blocks called amino acids (there are 20 kinds), which folds into a 3D shape. That shape decides what the protein can do, a bit like the shape of a key decides which lock it opens.
For decades, finding that shape took months or even years of experiments for a single protein. The DeepMind team points out that around 100,000 different structures had been determined, while billions of protein sequences are known. Since the 1970s, researchers had tried to calculate the shape directly, without a reliable result.
What did the DeepMind researchers do?
In 2020, Demis Hassabis and John Jumper presented AlphaFold2, a neural network (a program that learns from examples). It was built to include physical and biological knowledge about proteins, and it compares a protein's sequence with those of related proteins to spot which parts change together.
The team tested it in CASP14, an international competition that assesses methods for working out protein structures. The result was described in Nature in 2021 by John Jumper and colleagues.
What results convinced the Nobel committee?
According to the 2021 paper, this is the first computational method able to regularly find a protein's structure with atom-level accuracy, even when no similar structure is known. In most cases, the accuracy was comparable to structures obtained by experiments.
The Nobel committee's announcement says the tool has been used to work out the structure of virtually all of the 200 million proteins that researchers have identified, and that more than two million people from 190 countries have used it. The committee mentions uses such as better understanding antibiotic resistance and imaging enzymes that can break down plastic.
Key figures
- 200 millionknown proteins whose structure could be worked out with AlphaFold2, according to the Nobel committee
- more than 2 millionpeople in 190 countries have used AlphaFold2 since its release
- around 100,000protein structures determined by experiment before AlphaFold, against billions of known sequences
What does this change for patients?
No new treatment overnight, but a booster for research. When the shape of a protein involved in a disease is known, scientists can search for a molecule that fits it, or understand why a mutation disrupts it. Work that used to start with months of experiments can now start with a calculation.
This advance already feeds health projects, such as designing protein-based drugs or vaccines. But those projects still have to go through every usual step: lab tests, animal studies, then clinical trials.
What are the limits?
A calculated shape is still a model. The 2021 paper speaks of accuracy comparable to experimental structures in most cases, not all. Specialists therefore keep checking important structures in the lab.
Above all, knowing a protein's shape does not tell the whole story: how it moves, who it teams up with and what it triggers inside a cell are other questions, and any medicines that come out of this knowledge remain to be proven.
From research to your daily life: Mima
AlphaFold helps researchers understand proteins. It speeds up research but is not a treatment, and any medicines that come from it will take years.
In the meantime, Mima is here for daily life. She explains treatments and side effects in plain words and prepares your questions for your doctor. A question to ask: "Can you explain simply how my treatment works?"
Frequently asked questions
What is AlphaFold?
AlphaFold is an artificial intelligence from Google DeepMind that works out the 3D shape of a protein from its sequence of amino acids. Version 2 was presented in 2020 and described in Nature in 2021.
Who won the 2024 Nobel Prize in Chemistry?
The prize was shared: one half to David Baker (designing new proteins) and the other half to Demis Hassabis and John Jumper (AlphaFold).
Can AlphaFold cure diseases?
Not directly. The tool gives the shape of proteins, which helps researchers understand diseases and look for drugs. Any treatment must then be tested in clinical trials.
Can anyone use AlphaFold?
The Nobel committee says more than two million people in 190 countries have already used it, which shows it is widely available to researchers.
The original study
Callaway E. “Chemistry Nobel goes to developers of AlphaFold AI that predicts protein structures.” Nature, October 2024. DOI : 10.1038/d41586-024-03214-7
Read the full news story in NatureThis article summarises a scientific study and is not medical advice.