|
Meta predicts shape of 600 million proteins
Researchers at Meta (formerly Facebook) have used artificial intelligence (AI) to predict the structures of some 600 million proteins from bacteria, viruses and other microbes. Meta’s system, called ESMFold, isn’t quite as accurate as AlphaFold (DeepMind’s pioneering protein-structure AI), but it is about 60 times faster at predicting structures. Meta researchers tested their model on a ‘metagenomic’ database: DNA sequenced in bulk from environmental sources, including soil, seawater, the human gut and skin. The team predicted the structures of more than 617 million proteins, millions of which were entirely new to science. Such analyses “should cover a large part of the previously unseen protein universe”, says computational biologist Martin Steinegger. “There’s a big opportunity now to unravel more of the darkness.” Nature | 5 min read |
