“It’s a more complex system by quite a bit,” Jumper says. They also set it a more difficult, task: instead of predicting relationships between amino acids, the network predicts the final structure of a target protein sequence. So it changed tack, says Jumper, and developed an AI network that incorporated additional information about the physical and geometric constraints that determine how a protein folds.
The team tried to build on that approach but eventually hit the wall. In a second step that does not invoke AI, AlphaFold uses this information to come up with a ‘consensus’ model of what the protein should look like, says John Jumper at DeepMind, who is leading the project. The first iteration of AlphaFold applied the AI method known as deep learning to structural and genetic data to predict the distance between pairs of amino acids in a protein. But its approach was broadly similar to those of other teams that were applying AI, says Jinbo Xu, a computational biologist at the University of Chicago, Illinois. “You’re really finding out what looks promising, what works, and what you should walk away from,” he says.ĭeepMind’s 2018 performance at CASP13 startled many scientists in the field, which has long been the bastion of small academic groups. Moult credits the experiment - he doesn’t call it a competition - with vastly improving the field, by calling time on overhyped claims. The event challenges teams to predict the structures of proteins that have been solved using experimental methods, but for which the structures have not been made public. Moult started CASP to bring more rigour to these efforts. Lofty claims for methods in published papers tended to disintegrate when other scientists applied them to other proteins. Early attempts to use computers to predict protein structures in the 1980s and 1990s performed poorly, say researchers. Scientists have long wondered how a protein’s constituent parts - a string of different amino acids - map out the many twists and folds of its eventual shape. But, over the past decade, cryo-EM has become the favoured tool of many structural-biology labs. X-ray crystallography has produced the lion’s share of protein structures. The first complete structures of proteins were determined, starting in the 1950s, using a technique in which X-ray beams are fired at crystallized proteins and the diffracted light translated into a protein’s atomic coordinates. Proteins tend to adopt their shape without help, guided only by the laws of physics.įor decades, laboratory experiments have been the main way to get good protein structures.
How a protein works and what it does is determined by its 3D shape - ‘structure is function’ is an axiom of molecular biology. Proteins are the building blocks of life, responsible for most of what happens inside cells. AlphaFold might not obviate the need for these laborious and expensive methods - yet - say scientists, but the AI will make it possible to study living things in new ways. In some cases, AlphaFold’s structure predictions were indistinguishable from those determined using ‘gold standard’ experimental methods such as X-ray crystallography and, in recent years, cryo-electron microscopy (cryo-EM). AlphaFold has already helped him find the structure of a protein that has vexed his lab for a decade, and he expects it will alter how he works and the questions he tackles. “It’s a game changer,” says Andrei Lupas, an evolutionary biologist at the Max Planck Institute for Developmental Biology in Tübingen, Germany, who assessed the performance of different teams in CASP. But, this year, the outfit’s deep-learning network was head-and-shoulders above other teams and, say scientists, performed so mind-bogglingly well that it could herald a revolution in biology. It would vastly accelerate efforts to understand the building blocks of cells and enable quicker and more advanced drug discovery.ĪlphaFold came top of the table at the last CASP - in 2018, the first year that London-based DeepMind participated. The ability to accurately predict protein structures from their amino-acid sequence would be a huge boon to life sciences and medicine. AI protein-folding algorithms solve structures faster than ever