{"id":69488,"date":"2022-11-02T18:08:12","date_gmt":"2022-11-02T16:08:12","guid":{"rendered":"https:\/\/forklog.com\/en\/?p=69488"},"modified":"2025-09-07T20:20:27","modified_gmt":"2025-09-07T17:20:27","slug":"meta-ai-builds-a-competitor-to-deepminds-alphafold","status":"publish","type":"post","link":"https:\/\/forklog.com\/en\/meta-ai-builds-a-competitor-to-deepminds-alphafold\/","title":{"rendered":"Meta AI builds a competitor to DeepMind&#8217;s AlphaFold"},"content":{"rendered":"<p>Researchers at Meta AI released the &#8216;protein language model&#8217; ESM-2 with 15 billion parameters and the ESM Metagenomic Atlas database, containing over 600 million predicted structures of <span data-descr=\"collection of genetic material of microorganisms obtained directly from the environment\" class=\"old_tooltip\">metagenomic<\/span> compounds.<\/p>\n<figure class=\"wp-block-embed is-type-rich is-provider-twitter wp-block-embed-twitter\">\n<div class=\"wp-block-embed__wrapper\">\n<blockquote class=\"twitter-tweet\" data-width=\"500\" data-dnt=\"true\">\n<p lang=\"en\" dir=\"ltr\">Announcing the ESM Metagenomic Atlas \u2014 the first comprehensive view of the \u2018dark matter\u2019 of the protein universe. Made possible by ESMFold, a new breakthrough model for protein folding from Meta AI.<\/p>\n<p>More in our new blog \u27a1\ufe0f <a href=\"https:\/\/t.co\/LsUhSjzqCf\">https:\/\/t.co\/LsUhSjzqCf<\/a><\/p>\n<p>1\/3 <a href=\"https:\/\/t.co\/5lq48rPv5A\">pic.twitter.com\/5lq48rPv5A<\/a><\/p>\n<p>\u2014 AI at Meta (@AIatMeta) <a href=\"https:\/\/twitter.com\/AIatMeta\/status\/1587467591068459008?ref_src=twsrc%5Etfw\">November 1, 2022<\/a><\/p><\/blockquote>\n<p><script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script>\n<\/div>\n<\/figure>\n<p>Proteins are complex molecules made up of up to 20 amino acids and perform all kinds of biological functions in organisms. They fold into complex three-dimensional structures, the shapes of which directly influence how they work.<\/p>\n<p>Determining the type of interactions helps scientists understand how proteins function. Shape data also helps them find ways to imitate, modify or counter this behavior.<\/p>\n<p>You cannot derive the final structure from amino acid formulas alone, and simulations and experiments take a long time.<\/p>\n<p>In a statement, <a href=\"https:\/\/forklog.com\/en\/news\/what-are-transformers-machine-learning\">neural network transformer<\/a> ESM-2 is a large language model designed to \u201cstudy evolutionary patterns and generate accurate predictions of interactions directly from a protein sequence.\u201d<\/p>\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"576\" src=\"https:\/\/forklog.com\/wp-content\/uploads\/310840354_1952087951661612_1909789843050316859_n-1024x576.png\" alt=\"Meta AI develops a competitor to DeepMind's AlphaFold\" class=\"wp-image-189641\" srcset=\"https:\/\/forklog.com\/wp-content\/uploads\/310840354_1952087951661612_1909789843050316859_n-1024x576.png 1024w, https:\/\/forklog.com\/wp-content\/uploads\/310840354_1952087951661612_1909789843050316859_n-300x169.png 300w, https:\/\/forklog.com\/wp-content\/uploads\/310840354_1952087951661612_1909789843050316859_n-768x432.png 768w, https:\/\/forklog.com\/wp-content\/uploads\/310840354_1952087951661612_1909789843050316859_n-1536x864.png 1536w, https:\/\/forklog.com\/wp-content\/uploads\/310840354_1952087951661612_1909789843050316859_n-2048x1152.png 2048w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption>Protein structure prediction by a language model. Data: Meta AI.<\/figcaption><\/figure>\n<p>The system processes gene sequences using a self-supervised learning method called <a href=\"https:\/\/arxiv.org\/pdf\/2104.06644.pdf\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">masked language modelling<\/a>.<\/p>\n<p>According to researchers, they trained the algorithm on a dataset of sequences from millions of natural proteins.<\/p>\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cWith this approach, the model should correctly fill in words in a snippet of text, for example \u2018To __ or not __, that is __\u2019. We trained the language model to fill in gaps in protein sequences like \u2018GL_KKE_AHY_G\u2019 among millions of different interactions,\u201d the study says.<\/p>\n<\/blockquote>\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"576\" src=\"https:\/\/forklog.com\/wp-content\/uploads\/312062054_824222298698283_7847738217102467171_n-1024x576.png\" alt=\"ESM-2 fills gaps in protein sequences. Data: Meta AI.\" class=\"wp-image-189640\" srcset=\"https:\/\/forklog.com\/wp-content\/uploads\/312062054_824222298698283_7847738217102467171_n-1024x576.png 1024w, https:\/\/forklog.com\/wp-content\/uploads\/312062054_824222298698283_7847738217102467171_n-300x169.png 300w, https:\/\/forklog.com\/wp-content\/uploads\/312062054_824222298698283_7847738217102467171_n-768x432.png 768w, https:\/\/forklog.com\/wp-content\/uploads\/312062054_824222298698283_7847738217102467171_n-1536x864.png 1536w, https:\/\/forklog.com\/wp-content\/uploads\/312062054_824222298698283_7847738217102467171_n-2048x1152.png 2048w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption>ESM-2 fills gaps in protein sequences. Data: Meta AI.<\/figcaption><\/figure>\n<p>ESM-2 is the largest and most capable neural network of its kind. Scientists say the algorithm is 60 times faster than other contemporary systems such as AlphaFold from DeepMind.<\/p>\n<p>The algorithm helped create ESM Metagenomic Atlas, predicting 617 million structures from the protein database <a href=\"https:\/\/www.ebi.ac.uk\/metagenomics\/\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">MGnify90<\/a> in just two weeks on a cluster of 2,000 GPUs. For modelling a 384\u2011amino\u2011acid chain on one Nvidia V100 GPU, it would take about 14.2 seconds.<\/p>\n<figure class=\"wp-block-video\"><video controls src=\"https:\/\/forklog.com\/wp-content\/uploads\/313218089_148094367638525_8887962018934213013_n.mp4\"><\/video><\/figure>\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cWith current computing tools, predicting the structure of hundreds of millions of proteins could take years, even with the resources of a major research institution. To make predictions at the metagenomics scale, a breakthrough in speed is crucial,\u201d the developers noted.<\/p>\n<\/blockquote>\n<p>Meta AI hopes that ESM-2 and the ESM Metagenomic Atlas will advance science and help researchers studying evolutionary history or combating disease and climate change.<\/p>\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cWe are also exploring ways to apply language models to develop new proteins and help address health and environmental problems,\u201d the scientists added.<\/p>\n<\/blockquote>\n<p>In July, DeepMind&#8217;s AlphaFold predicted <a href=\"https:\/\/forklog.com\/en\/news\/alphafold-predicts-nearly-all-known-proteins\">almost all known biological compounds<\/a> discovered in plants, bacteria and animals.<\/p>\n<p>In the same month, MIT researchers <a href=\"https:\/\/forklog.com\/en\/news\/ai-can-find-drug-candidates-1200-times-faster-than-current-systems\">developed the deep-learning model EquiBind<\/a>, which binds molecules to proteins for drug design 1,200 times faster than its rivals.<\/p>\n<p>In July 2021, DeepMind&#8217;s AI <a href=\"https:\/\/forklog.com\/en\/news\/deepmind-ai-maps-20000-human-protein-structures\">modeled 20,000 human protein structures<\/a>.<\/p>\n<p>Subscribe to ForkLog&#8217;s Telegram news: <a href=\"https:\/\/t.me\/forklogAI\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">ForkLog AI<\/a> \u2014 all the news from the world of AI!<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Researchers at Meta AI released the &#8216;protein language model&#8217; ESM-2 with 15 billion parameters and the ESM Metagenomic Atlas database, containing more than 600 million predicted structures of metagenomic compounds.<\/p>\n","protected":false},"author":1,"featured_media":69489,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"select":"1","news_style_id":"1","cryptorium_level":"","_short_excerpt_text":"","creation_source":"","_metatest_mainpost_news_update":false,"footnotes":""},"categories":[3],"tags":[438,1414,1293,167],"class_list":["post-69488","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news-and-analysis","tag-artificial-intelligence","tag-medicine","tag-meta","tag-research"],"aioseo_notices":[],"amp_enabled":true,"views":"45","promo_type":"1","layout_type":"1","short_excerpt":"","is_update":"","_links":{"self":[{"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/posts\/69488","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/comments?post=69488"}],"version-history":[{"count":1,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/posts\/69488\/revisions"}],"predecessor-version":[{"id":69490,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/posts\/69488\/revisions\/69490"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/media\/69489"}],"wp:attachment":[{"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/media?parent=69488"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/categories?post=69488"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/tags?post=69488"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}