The race to crack one of biology’s most significant challenges — predicting the 3D structures of proteins from their amino-acid sequences — is intensifying, due to new artificial-intelligence (AI) approaches.
On the end of last year, Google’s AI firm DeepMind debuted an algorithm referred to as AlphaFold, which combined two methods that had been emerging within the field and beat established contenders in a competition on the protein-structure prediction by a surprising margin. And in April this year, a US researcher revealed an algorithm that uses a very different method. He claims his AI is up to a million times faster at predicting buildings than DeepMind’s, although probably not as accurate in all situations.
More broadly, biologists are questioning how else deep learning — the AI technique utilized by both approaches — is perhaps applied to the prediction of protein preparations, which ultimately dictate a protein’s function. These approaches are cheaper and faster than existing lab techniques such as X-ray crystallography, and the data might assist researchers in understanding diseases and design medication better. “There’s a lot of excitement about where things might go now,” says John Moult, a biologist at the University of Maryland in College Park and the founder of the biennial competition, referred to as Critical Assessment of protein Structure Prediction (CASP), where groups are challenged to design computer programs that predict protein structures from sequences.
The newest algorithm’s creator, Mohammed AlQuraishi, a biologist at Harvard Medical School in Boston, Massachusetts, hasn’t but directly compared the accuracy of his technique with that of AlphaFold — and he suspects that AlphaFold would beat his method inaccuracy when proteins with sequences similar to the one being analyzed are available for reference. However, he says that as a result of his algorithm uses a mathematical function to calculate protein structures in a single step — rather than in two steps like AlphaFold, which uses the same structures as groundwork in the first step — it can predict structures in milliseconds rather than hours or days.