Predicting the shape

Section 5 of 5

Reading the order of a protein's amino acids from its gene is quick. Finding its fold by experiment once took years. Computers now predict the fold from the order alone.

Seeing a fold

The classic way to see a fold is X-ray crystallography. Chemists coax the protein into a crystal, shine X-rays through it, and record the pattern of spots the X-rays make. From the pattern they work back to where every atom sits. In 1958 John Kendrew used it to find the first protein fold, myoglobin, which stores oxygen in muscle. Small proteins can also be solved with NMR, the magnet method that reads hydrogen atoms, and since the 2010s a third method, cryo-electron microscopy, has photographed many thousands of frozen protein molecules with a beam of electrons and combined the pictures into one 3D fold.

Every fold found goes into the Protein Data Bank, which opened in 1971 with 7 of them and now holds more than 250,000. Gene reading finds new proteins far faster: hundreds of millions of sequences are known.

Too many folds to try

Working a fold out from the sequence alone is hard. Suppose each amino acid could sit in just 3 positions. A chain of 100 would then have 3 multiplied by itself 100 times: a number of possible folds 48 digits long. Trying a million million of them every second would take about a billion billion times the age of the universe. Small proteins can fold in under a second, so a chain never tries them all: as the oily core gathers, each step leads toward the right fold, like a ball rolling downhill.

Since 1994, a contest called CASP has run every two years. Teams predict the folds of proteins whose shapes have just been found in the lab, before those shapes are published, and their predictions are scored against the real ones. In 2020 AlphaFold 2, from DeepMind, a Google company, took part, and for about two thirds of the proteins its predictions came close to the precision of the experiments.

Finding protein shapes, from 1958 to 2024 A timeline running down the page. 1958: myoglobin, the first protein shape, by X-ray crystallography. 1971: the Protein Data Bank opens, with 7 shapes. 1994: the first CASP contest to predict shapes. 2003: Top7, a new fold designed on a computer. 2020: AlphaFold 2 comes close to the precision of experiments. 2022: predicted shapes for over 200 million proteins. 2024: the Nobel Prize for Baker, Hassabis and Jumper. The last three dots are amber. 1958 Myoglobin: the first protein shape, by X-ray crystallography 1971 The Protein Data Bank opens, with 7 shapes 1994 The first CASP contest to predict shapes 2003 Top7, a new fold designed on a computer 2020 AlphaFold 2 comes close to the precision of experiments 2022 Predicted shapes for over 200 million proteins 2024 Nobel Prize: Baker, Hassabis and Jumper
Finding protein shapes, from the first fold seen by X-rays to the 2024 Nobel Prize. The last three steps came in just four years.

AlphaFold learned from the folds already in the Protein Data Bank. It also compares a sequence with the same protein in many other living things: two amino acids that tend to change together over evolution usually sit next to each other in the fold. In 2022 its makers and the European Bioinformatics Institute released predicted folds for over 200 million proteins, free for anyone to use.

David Baker's team, at the University of Washington, works the other way round. Their programs design a fold first and then find a sequence that will take it. In 2003 they made Top7, a protein with a fold never seen in nature, and it folded just as the computer had planned.

Quick check

How did AlphaFold learn to predict a protein's fold from its sequence?

That's the end of lesson 46

In this lesson you found out:

  • How the order of amino acids decides a protein's fold, and what holds it
  • What heat, acid and salt do to a protein's fold
  • What happens when proteins fold the wrong way
  • How computers predict a protein's fold from its sequence

Saved

Sort saved items

Starred items stay on top in either order.

Save a figure, a table or a section with its bookmark button, and it appears here.