Google’s AI subsidiary DeepMind announced earlier this year AlphaFold, an AI system for predicting protein structures. Now researchers at social networking giant Meta/Facebook have done similar work, using a language model trained on sequences of known protein structures and then using it to predict the structures of other proteins. The researchers call their AI model ESMFold, which is not as accurate as AlphaFold, but predicts protein structures 60 times faster than AlphaFold. It took the researchers two weeks to predict the structures of more than 617 million proteins, compared to the minutes it took AlphaFold to complete a single protein structure prediction. Of the 617 million structures, more than a third were considered high quality and considered correct. Millions of them are completely new and different from databases of known protein structures.
This article is reprinted from: https://www.solidot.org/story?sid=73246
This site is for inclusion only, and the copyright belongs to the original author.