What A Time To Be Alive

Glossary

Protein folding

Protein folding is the process by which a chain of amino acids folds into a three-dimensional shape. That shape determines what the protein can do. AI can now often predict that shape well.

A protein is made in the cell as a long, loose chain of amino acids. Such a chain can do nothing. Only once it has folded into a three-dimensional shape does the protein do its work. That folding is called protein folding.

How does it work?

The order of the amino acids determines how the chain folds. Some amino acids attract water, others avoid it. Some attract each other, others repel each other. Through all these forces, the chain falls in a short time into the shape in which it is most stable, a bit like an elastic band that settles into a tangle by itself.

Calculating which shape that will be was one of the big open questions of biology for decades. A long chain can fold in an unimaginable number of ways. Measuring a shape in the lab also takes a great deal of time.

For some years now, AI models such as Google DeepMind's AlphaFold have been able to predict the shape of many proteins from the sequence of amino acids alone. For that work, Demis Hassabis and John Jumper received the 2024 Nobel Prize in Chemistry, together with David Baker, who is known for his work on designing proteins. A prediction remains a prediction: researchers still check it against measurements.

Sometimes the folding goes wrong. A misfolded protein cannot do its job or clumps together. That plays a role in diseases such as Alzheimer's.

Why it matters

Anyone who knows the shape of a protein can understand how it works and how a medicine can attach to it. Movement counts too, because a protein does not stand still. That is why Microsoft released BioEmu-2, a model that lets researchers predict which shapes a protein or other biomolecule can take, including together with DNA, RNA or small molecules. The model is free and meant for research (link).

Folding also plays a part in other molecules. Researchers from China presented structRFM in Nature, a model that learns how RNA molecules fold. According to its makers, it keeps up with Google DeepMind's AlphaFold3 in tests, and the data, model and code are freely available (link).

Articles about Protein folding