Artificial intelligence is once again in the spotlight in this week of Nobel Prize announcements. The chemistry section of the promotion, awarded on Wednesday, October 9, rewards, at least half, an artificial intelligence tool that has revolutionized the lives of biochemists, AlphaFold. The British Demis Hassabis (48 years old) and the American John Jumper (39 years old) developed this software within the company Deepmind, co-founded by the former in 2010 and bought by Google in 2014. The other half of the price goes to the American David Baker (62 years old), from the University of Washington, who twenty years before his co-laureates had tackled the same topic, without artificial intelligence but with initial success.
This problem is a kind of enigma, difficult for humans to solve, but easy for Mother Nature: how to find the three-dimensional shape that proteins adopt knowing only the sequence of the twenty amino acids that constitute them? A virus attaches to a cell using proteins (like the famous coronavirus spike). Iron is transported in the blood by proteins (hemoglobin). Hormones and enzymes are proteins…
Formed by a string of pearls, amino acids and proteins do not stay aligned like spaghetti, but rather they fold, twist, form helices, hooks, bags… which in the end give them their function. Pathologies such as Creutzfeldt-Jakob disease are even related to defects in protein folding.
However, the information available to chemists about them is often a sequence of genes that code for amino acids and are therefore assembled into proteins. But this is not enough to understand its function or its mode of action. The form in space is necessary. And it is this thorny problem that the Deepmind team, around thirty people, led by John Jumper, managed to solve in a work published in July 2021, using various machine learning techniques using artificial neural networks.
“Everyone uses AlphaFold!” »
A few months earlier, at the end of 2020, during a competition involving several teams to find structures from sequences, version 2 of AlphaFold crushed the competition, doubling the performance obtained until then. Deepmind then details its algorithm, publishes its code online, and announces its collaboration with the European Molecular Biology Laboratory (EMBL) to enrich a database whose information was used to develop the software.
You have 64.25% of this article left to read. The rest is reserved for subscribers.
