Researchers at Meta claim to have created a massive model of protein folding that has predicted the structure of 617 million proteins. Meta said its database predicts proteins found in microbes in the soil, deep in the ocean and inside human bodies. The company said these proteins are some of the least understood on Earth. The research team published their results in preprint. The team said it made the predictions using a large language model, which is an artificial intelligence system trained on massive amounts of data to solve math problems and generate text. Meta said it trained the large language model “to learn evolutionary patterns and generate accurate end-to-end structure predictions directly from a protein’s sequence.” The company claims that this ESMFold model’s predictions are 60 times faster than other state-of-the-art systems, allowing for larger databases. In total, the team said their model was trained on 15 billion parameters and was able to predict more than 617 million protein structures over two weeks. The company said it shared that database, called the ESM Metagenomic Atlas, along with an API to allow scientists to “easily retrieve specific protein structures relevant to their work.” “ESMFold shows how artificial intelligence can give us new tools to understand the natural world, like the microscope, which has enabled us to see the world at the infinitesimal scale and opened up a whole new understanding of life,” said Go to a blog post. . If Meta’s AI model proves accurate, it could rival AlphaFold, the AI developed by Google-owned DeepMind. Earlier this year, DeepMind said its AlphaFold AI made a huge scientific breakthrough by predicting the structure of more than 200 million proteins. The company first shared details of AlphaFold in late 2020, when it claimed its AI system had solved the “protein folding” problem. A year ago, AlphaFold was released as an open source project and the company created the AlphaFold Protein Structure Database. 10 things you need to know straight to your inbox every day. Sign up for the Daily Brief, Silicon Republic’s digest of key tech science news.
