AI unravels 50-year-old biological riddle

This will change everything: T1037, part of a protein from a virus that infects bacteria © DeepMind

Was this the most important news story of 2020? Researchers at DeepMind have made a mind-boggling discovery that could eradicate diseases and put the brakes on climate breakdown.

By 2 December 2030, the world’s landfill problem might have been solved. Terrible illnesses like Alzheimer’s and Parkinson’s disease could be as much a thing of the past as smallpox and polio. And it will all be thanks to a scientific breakthrough unveiled on Monday.

It seems like a very small, technical achievement: the artificial intelligence group DeepMind has produced a programme, AlphaFold, that can reliably predict how proteins will fold. But it will revolutionise the way humanity lives.

All life is based on proteins, from the tiniest virus to the biggest, most complex creature. Being able to map proteins is the key to all kinds of revolutionary technologies.

We could produce “green enzymes” that safely break down plastic waste. We could predict and prevent cases of neurodegenerative diseases like Huntington’s disease. We could even cure age-old illnesses like malaria, which has plagued humanity since the beginning of our history.

But what is protein folding, and why is it so important?

A gene is a long string of DNA molecules which are described using four letters: c, a, t and g. Three of these letters together make up an amino acid. If the molecules are like letters in a word, then amino acids are the words themselves.

But just knowing the “words” does not help us understand the proteins. Proteins are clumps of amino acids that are folded in particular ways, and these folds are like grammar: they determine the function of the protein in the same way that the order of words decides the meaning of a sentence.

The two sentences “the dog bites the man” and “the man bites the dog” have the same words, but very different meanings. In the same way, identical amino acids can produce very different proteins.

Genetic disorders are caused when a single DNA molecule is “misprinted” as another letter: a kind of genetic typo. This single, tiny mutation can alter the way in which a protein folds, with devastating effects on the body.

Previously, scientists could map out a protein. But they could not predict how a mutation would affect its shape, and thus its function. It is a problem that has plagued biology for five decades. Now they finally have a solution. It is like cracking the Enigma code in the fight against genetic disease.

Longer life expectancies across the world have come at a price: genetic disorders associated with old age – like Alzheimer’s and Parkinson’s – which occur when brain proteins fold in the wrong way. Now we have the key to living longer lives without the threat of brain degeneration.

The breakthrough is also a vindication for DeepMind, which has previously focused mostly on training AI to play games like chess and Go. This is the long-term strategy of the Alphabet-owned company: it has used games as a training ground for its AI, building them up to tackle more serious problems.

There is more work to be done: AlphaFold is not correct 100% of the time. But it suggests that AI does have an application in this vital area of science.

So, is this the most important story of 2020?

Protein shake

Absolutely, say some. This is a huge leap forward for science, with plenty of practical applications. Thanks to this discovery, the world could be unrecognisable by the beginning of the next decade. Whole diseases might have been wiped out, the oceans could be free of plastic, famine could be a thing of the past, and old age might no longer carry the threat of slow, agonising mental decline.

Hold your horses, say others. While this is a huge breakthrough, it will take time and a lot more experimentation to put it into practice. While DeepMind’s AI can predict protein folding, we still do not understand how it works: if anything, the mystery has deepened. Science is a continuous process, and it will take more work and diligent research before the new programme starts changing our lives.

You Decide

  1. AlphaFold can “learn” independently of human beings, by studying data. Are you worried that AI might become too sophisticated for humans to control? Or should we always push for progress, regardless of the risks?
  2. DeepMind has not yet released information on the workings of the AlphaFold programme. Should they be required to divulge this information for others to scrutinise? Or are they entitled to their industrial secrets?


  1. Write down the letters c, a, t and g three times, then do it again in the line below with the order changed. See how many variations of the 12 letters you can write.
  2. Imagine you are part of the DeepMind team that developed AlphaFold. Write a diary entry describing how you feel about the success of the programme.

Some People Say...

“Our universe is a sorry little affair unless it has in it something for every age to investigate.”

Seneca (4BC - 65AD), Roman philosopher

What do you think?

Q & A

What do we know?
Most people agree that scientific advance comes on in leaps and bounds. It is common to talk of science in terms of revolutions. The Copernican revolution changed our understanding of the shape of the solar system, after Copernicus showed that the Earth revolves around the sun. The Newtonian revolution made us think about the world in terms of physical forces. The philosopher of science Thomas Kuhn argued that all science progresses as a result of these “paradigm shifts”.
What do we not know?
There is debate over how far AI can advance our actual understanding. DeepMind’s AlphaFold programme can reliably predict how a protein will develop. The problem is that we are not sure how AlphaFold knows that. It is considered a “black box” technology: its inner workings are unclear. The whole idea of AI is that it should be able to learn without informational input from human beings, but it means that for every mystery it solves, AI may create yet another mystery for human beings.

Word Watch

Landfill problem
Every year, human beings create more than two billion tonnes of waste. Most of this ends up being dumped in vast landfill sites, where it produces the greenhouse gas methane, or in the ocean, where it harms marine life.
A UK-based artificial intelligence developer, founded in 2010 and acquired by Google in 2014 for £400m.
A kind of molecule that drives all biological processes. They make up around one-fifth of our total body mass and without them our bodies could not function. They fold into 3D shapes, making them extraordinarily complicated as a whole.
Neurodegenerative diseases
Illnesses that affect the nervous system, causing brain function to decrease and often resulting in motor failures as well.
This often fatal disease is carried by mosquitoes, which tend to gather around standing water. The development of agriculture caused an explosion in their numbers, and malaria became a corollary to human civilisation.
A typing mistake in a piece of writing
A coding device used by Germany in World War Two, thought to be impossible to crack. It was first decoded by Polish engineers in the 1930s. In 1941, Alan Turing developed a machine that was capable of cracking even its tightest codes.
The oldest board game still played in the world. It originated in China and is still most popular there. Its rules are simple but in reality it is very complicated: there are more legal board positions in the game than there are atoms in the known universe.
The parent company of Google and an array of other subsidiaries.

PDF Download

Please click on "Print view" at the top of the page to see a print friendly version of the article.