{"id":3926,"date":"2019-07-22T20:55:21","date_gmt":"2019-07-22T11:55:21","guid":{"rendered":"http:\/\/163.180.4.222\/lab\/?p=3926"},"modified":"2019-07-22T20:55:21","modified_gmt":"2019-07-22T11:55:21","slug":"ai-protein-folding-algorithms-solve-structures-faster-than-ever","status":"publish","type":"post","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=3926","title":{"rendered":"AI protein-folding algorithms solve structures faster than ever"},"content":{"rendered":"<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<h5>Deep learning makes its mark on protein-structure prediction.<\/h5>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<figure class=\"figure\">\n<div class=\"embed intensity--high\">\n<div class=\"embed intensity--high\"><img decoding=\"async\" class=\"figure__image\" src=\"https:\/\/media.nature.com\/w800\/magazine-assets\/d41586-019-01357-6\/d41586-019-01357-6_16964586.jpg\" alt=\"Prokaryotes and eukaryotes respond to heat shock and other forms of environmental stress\" data-src=\"\/\/media.nature.com\/w800\/magazine-assets\/d41586-019-01357-6\/d41586-019-01357-6_16964586.jpg\" \/><\/div>\n<\/div><figcaption>\n<p class=\"figure__caption sans-serif\"><span class=\"mr10\">Predicting protein structures from their sequences would aid drug design.<\/span>Credit: Edward Kinsman\/Science Photo Library<\/p>\n<\/figcaption><\/figure>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<p>The race to crack one of biology\u2019s grandest challenges \u2014 predicting the 3D structures of proteins from their amino-acid sequences \u2014 is intensifying, thanks to new artificial-intelligence (AI) approaches.<\/p>\n<p>At the end of last year, Google\u2019s AI firm DeepMind debuted an algorithm called\u00a0<a href=\"https:\/\/deepmind.com\/blog\/alphafold\/\" data-track=\"click\" data-label=\"https:\/\/deepmind.com\/blog\/alphafold\/\" data-track-category=\"body text link\">AlphaFold<\/a>, which combined two techniques that were emerging in the field and beat established contenders in a competition on protein-structure prediction by a surprising margin. And in April this year, a US researcher revealed an algorithm that uses a totally different approach. He claims his AI is up to one million times faster at predicting structures than DeepMind\u2019s, although probably not as accurate in all situations.<\/p>\n<p>&nbsp;<\/p>\n<aside class=\"recommended pull pull--left sans-serif\" data-label=\"Related\"><a href=\"https:\/\/www.nature.com\/news\/machine-learning-predicts-the-look-of-stem-cells-1.21769\" data-track=\"click\" data-track-label=\"recommended article\"><img decoding=\"async\" class=\"recommended__image\" src=\"https:\/\/media.nature.com\/w400\/magazine-assets\/d41586-019-01357-6\/d41586-019-01357-6_15338610.jpg\" \/><\/a><\/p>\n<p class=\"recommended__title serif\">Machine learning predicts the look of stem cells<\/p>\n<\/aside>\n<p>&nbsp;<\/p>\n<p>More broadly, biologists are wondering how else deep learning \u2014 the AI technique used by both approaches \u2014 might be applied to the prediction of protein arrangements, which ultimately dictate a protein\u2019s function. These approaches are cheaper and faster than\u00a0<a href=\"https:\/\/www.nature.com\/news\/the-revolution-will-not-be-crystallized-a-new-method-sweeps-through-structural-biology-1.18335\" data-track=\"click\" data-label=\"https:\/\/www.nature.com\/news\/the-revolution-will-not-be-crystallized-a-new-method-sweeps-through-structural-biology-1.18335\" data-track-category=\"body text link\">existing lab techniques such as X-ray crystallography<\/a>, and the knowledge could help researchers to better understand diseases and design drugs. \u201cThere\u2019s a lot of excitement about where things might go now,\u201d says John Moult, a biologist at the University of Maryland in College Park and the founder of the biennial competition, called Critical Assessment of protein Structure Prediction (CASP), where teams are challenged to design computer programs that predict protein structures from sequences.<\/p>\n<p>&nbsp;<\/p>\n<p><strong>Innovative approach<\/strong><\/p>\n<p>The latest algorithm\u2019s creator, Mohammed AlQuraishi, a biologist at Harvard Medical School in Boston, Massachusetts, hasn\u2019t yet directly compared the accuracy of his method with that of AlphaFold \u2014 and he suspects that AlphaFold would beat his technique in accuracy when proteins with sequences similar to the one being analysed are available for reference. But he says that because his algorithm uses a mathematical function to calculate protein structures in a single step \u2014 rather than in two steps like AlphaFold, which uses the similar structures as groundwork in the first step \u2014 it can predict structures in milliseconds rather than hours or days.<\/p>\n<p>\u201cAlQuraishi\u2019s approach is very promising. It builds on advances in deep learning as well as some new tricks AlQuraishi has invented,\u201d says Ian Holmes, a computational biologist at the University of California, Berkeley. \u201cIt might be possible that, in the future, his idea can be combined with others to advance the field,\u201d says Jinbo Xu, a computer scientist at the Toyota Technological Institute at Chicago, Illinois, who competed at CASP13.<\/p>\n<p>At the core of AlQuraishi\u2019s system is a neural network, a type of algorithm inspired by the brain\u2019s wiring that learns from examples. It\u2019s fed with known data on how amino-acid sequences map to protein structures and then learns to produce new structures from unfamiliar sequences. The novel part of his network lies in its ability to create such mappings end-to-end; other systems use a neural network to predict certain features of a structure, then another type of algorithm to laboriously search for a plausible structure that incorporates those features. AlQuraishi\u2019s network takes months to train, but once trained, it can transform a sequence to a structure almost immediately.<\/p>\n<p>&nbsp;<\/p>\n<aside class=\"recommended pull pull--left sans-serif\" data-label=\"Related\"><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-02156-9\" data-track=\"click\" data-track-label=\"recommended article\"><img decoding=\"async\" class=\"recommended__image\" src=\"https:\/\/media.nature.com\/w400\/magazine-assets\/d41586-019-01357-6\/d41586-019-01357-6_16964784.jpg\" \/><\/a><\/p>\n<p class=\"recommended__title serif\">No limit: AI poker bot is first to beat professionals at multiplayer game<\/p>\n<\/aside>\n<p>&nbsp;<\/p>\n<p>His approach, which he dubs a recurrent geometric network, predicts the structure of one segment of a protein partly on the basis of what comes before and after it. This is similar to how people\u2019s interpretation of a word in a sentence can be influenced by surrounding words; these interpretations are in turn influenced by the focal word.<\/p>\n<p>Technical difficulties meant AlQuraishi\u2019s algorithm did not perform well at CASP13. But he has tested it on tests from previous published details of the AI in\u00a0<i>Cell Systems<\/i>\u00a0in April<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01357-6?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR1\" data-track=\"click\" data-action=\"anchor-link\" data-track-label=\"go to reference\" data-track-category=\"references\">1<\/a><\/sup>\u00a0and made his code publicly\u00a0<a href=\"https:\/\/github.com\/aqlaboratory\/proteinnet\" data-track=\"click\" data-label=\"https:\/\/github.com\/aqlaboratory\/proteinnet\" data-track-category=\"body text link\">available on GitHub<\/a>, hoping others will build on the work. (The structures for most of the proteins tested in CASP13 have not been made public yet, so he still hasn\u2019t been able to directly compare his method with AlphaFold.)<\/p>\n<p>&nbsp;<\/p>\n<p><strong>Neural networks<\/strong><\/p>\n<p>AlphaFold competed successfully at CASP13 and created a stir when it outperformed all other algorithms on hard targets by nearly 15%, according to one measure.<\/p>\n<p>AlphaFold works in two steps. Like other approaches used in the competition, it starts with something called multiple sequence alignments. It compares a protein\u2019s sequence with similar ones in a database to reveal pairs of amino acids that don\u2019t lie next to each other in a chain, but that tend to appear in tandem. This suggests that these two amino acids are located near each other in the folded protein. DeepMind trained a neural network to take such pairings and predict the distance between two paired amino acids in the folded protein.<\/p>\n<p>By comparing its predictions with precisely measured distances in proteins, it learnt to make better guesses about how proteins would fold up. A parallel neural network predicted the angles of the joints between consecutive amino acids in the folded protein chain.<\/p>\n<p>&nbsp;<\/p>\n<aside class=\"recommended pull pull--left sans-serif\" data-label=\"Related\"><a href=\"https:\/\/www.nature.com\/news\/the-revolution-will-not-be-crystallized-a-new-method-sweeps-through-structural-biology-1.18335\" data-track=\"click\" data-track-label=\"recommended article\"><img decoding=\"async\" class=\"recommended__image\" src=\"https:\/\/media.nature.com\/w400\/magazine-assets\/d41586-019-01357-6\/d41586-019-01357-6_16142182.jpg\" \/><\/a><\/p>\n<p class=\"recommended__title serif\">The revolution will not be crystallized: a new method sweeps through structural biology<\/p>\n<\/aside>\n<p>&nbsp;<\/p>\n<p>But these steps can\u2019t predict a structure by themselves, because the exact set of distances and angles predicted might not be physically possible. So in a second step, AlphaFold created a physically possible \u2014 but nearly random \u2014 folding arrangement for a sequence. Instead of another neural network, it used an optimization method called gradient descent to iteratively refine the structure so it came close to the (not-quite-possible) predictions from the first step.<\/p>\n<p>A few other teams used one of the approaches, but none used both. In the first step, most teams merely predicted contact in pairs of amino acids, not distance. In the second step, most used complex optimization rules instead of gradient descent, which is almost automatic.<\/p>\n<p>\u201cThey did a great job. They\u2019re about one year ahead of the other groups,\u201d says Xu.<\/p>\n<p>&nbsp;<\/p>\n<p><strong>Future Directions<\/strong><\/p>\n<p>DeepMind is yet to release all the details about AlphaFold \u2014 but other groups have since started adopting tacticsdemonstrated by DeepMind and other leading teams at CASP13. Jianlin Cheng, a computer scientist at the University of Missouri in Columbia, says he\u2019ll modify his deep neural networks to have some features of AlphaFold\u2019s, for instance by adding more layers to the neural network in distance-predicting stage. Having more layers \u2014 a deeper network \u2014 often allows networks to process information more deeply, hence the name deep learning.<\/p>\n<p>\u201cWe look forward to seeing similar systems put to use,\u201d says Andrew Senior, the computer scientist at DeepMind who led the AlphaFold team.<\/p>\n<p>Moult said there was a lot of discussion at CASP13 about how else deep learning might be applied to protein folding. Maybe it could help to refine approximate structure predictions; report on how confident the algorithm is in a folding prediction; or model interactions between proteins.<\/p>\n<p>And although computational predictions aren\u2019t yet accurate enough to be widely used in drug design, the increasing accuracy allows for other applications, such as understanding how a mutated protein contributes to disease or knowing which part of a protein to turn into a vaccine for immunotherapy. \u201cThese models are starting to be useful,\u201d Moult says.<\/p>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<p>(\uc6d0\ubb38: <a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01357-6?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29\">\uc5ec\uae30<\/a>\ub97c \ud074\ub9ad\ud558\uc138\uc694~)<\/p>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n","protected":false},"excerpt":{"rendered":"<p>&nbsp; &nbsp; Deep learning makes its mark on protein-structure prediction. &nbsp; &nbsp; Predicting protein structures from their sequences would aid drug design.Credit: Edward Kinsman\/Science Photo<a href=\"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=3926\" class=\"more-link\">(more&#8230;)<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"jetpack_post_was_ever_published":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[33,34,35,29,30],"tags":[],"class_list":["post-3926","post","type-post","status-publish","format-standard","hentry","category-do-biology","category-lets-do-chemistry","category-lets-do-computer-science","category-lets-do-science","category-recent-science-news"],"aioseo_notices":[],"jetpack_publicize_connections":[],"jetpack_featured_media_url":"","jetpack-related-posts":[{"id":2582,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=2582","url_meta":{"origin":3926,"position":0},"title":"Technologies to watch in 2019","author":"biochemistry","date":"January 29, 2019","format":false,"excerpt":"\u00a0 \u00a0 From higher-resolution imaging to genome-sized DNA molecules built from scratch, the year ahead looks exciting for life-science technology. \u00a0 An automated bioreactor system for growing yeast, which can be used to investigate synthetic genomes \u2014 one area poised to make big strides this year.Credit: Tim Llewellyn\/Ginkgo Bioworks \u00a0\u2026","rel":"","context":"In &quot;Essays on Science&quot;","block_context":{"text":"Essays on Science","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=32"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":4207,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=4207","url_meta":{"origin":3926,"position":1},"title":"The structure of DNA","author":"biochemistry","date":"October 11, 2019","format":false,"excerpt":"\u00a0 \u00a0 In the early 1950s, the identity of genetic material was still a matter of debate. The discovery of the helical structure of double-stranded DNA settled the matter \u2014 and changed biology forever. \u00a0 \u00a0 On 25 April 1953, James Watson and Francis Crick announced1\u00a0in\u00a0Nature\u00a0that they \u201cwish to suggest\u201d\u00a0a\u2026","rel":"","context":"In &quot;Essays on Science&quot;","block_context":{"text":"Essays on Science","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=32"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":3941,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=3941","url_meta":{"origin":3926,"position":2},"title":"The computational protein designers","author":"biochemistry","date":"July 27, 2019","format":false,"excerpt":"\u00a0 \u00a0 A new breed of protein engineers is finding that the best way to create a molecule is to build it from scratch. \u00a0 \u00a0 By designing a protein from the ground up, researchers can create molecules with forms and functions not found in nature.Credit: Brian DalBalcon \u00a0 \u00a0\u2026","rel":"","context":"In &quot;Let's Do Biology!&quot;","block_context":{"text":"Let's Do Biology!","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=33"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":4907,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=4907","url_meta":{"origin":3926,"position":3},"title":"Hunting for New Drugs with AI","author":"biochemistry","date":"December 19, 2019","format":false,"excerpt":"\u00a0 The pharmaceutical industry is in a drug-discovery slump. How much can AI help? \u00a0 \u00a0 Illustration by Harry Campbell \u00a0 \u00a0 There are many reasons that promising drugs wash out during pharmaceutical development, and one of them is cytochrome P450. A set of enzymes mostly produced in the liver,\u2026","rel":"","context":"In &quot;'06. \uc5d0\ub108\uc9c0\uc640 \uc5d4\ud2b8\ub85c\ud53c'\uc640 '07. \uacfc\ud559\uacfc \ubb38\uba85' \uad00\ub828&quot;","block_context":{"text":"'06. \uc5d0\ub108\uc9c0\uc640 \uc5d4\ud2b8\ub85c\ud53c'\uc640 '07. \uacfc\ud559\uacfc \ubb38\uba85' \uad00\ub828","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=42"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":4961,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=4961","url_meta":{"origin":3926,"position":4},"title":"Powerful antibiotics discovered using AI","author":"biochemistry","date":"February 24, 2020","format":false,"excerpt":"\u00a0 Machine learning spots molecules that work even against \u2018untreatable\u2019 strains of bacteria. \u00a0 \u00a0 Escherichia coli\u00a0bacteria, coloured green, in a scanning electron micrograph.Credit: Stephanie Schuller\/SPL \u00a0 \u00a0 A pioneering machine-learning approach has identified powerful new types of antibiotic from a pool of more than 100 million molecules \u2014 including\u2026","rel":"","context":"In &quot;Let's Do Biology!&quot;","block_context":{"text":"Let's Do Biology!","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=33"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":2987,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=2987","url_meta":{"origin":3926,"position":5},"title":"Precise packing for membrane proteins","author":"biochemistry","date":"March 29, 2019","format":false,"excerpt":"\u00a0 \u00a0 Although nonpolar amino acid side chains pack efficiently in membrane proteins, it has been difficult to determine how much this contributes to membrane protein stability. Designed membrane proteins have largely relied on other stabilizing interactions such as metal-ligand interactions and hydrogen bonds. Mravic\u00a0et al.\u00a0uncovered a steric packing code\u2026","rel":"","context":"In &quot;Let's Do Biology!&quot;","block_context":{"text":"Let's Do Biology!","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=33"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]}],"jetpack_sharing_enabled":false,"jetpack_shortlink":"https:\/\/wp.me\/p9Xo1j-11k","_links":{"self":[{"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/posts\/3926","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=3926"}],"version-history":[{"count":1,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/posts\/3926\/revisions"}],"predecessor-version":[{"id":3927,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/posts\/3926\/revisions\/3927"}],"wp:attachment":[{"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3926"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3926"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3926"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}