成功破解困擾生物學界50年的蛋白質摺疊難題，DeepMind的AlphaFold 2已宣佈開源！

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"蛋白質結構預測技術正在走向大衆。廣大科學家很快就能用上準確預測蛋白質3D形狀的軟件。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"7月15日，總部位於倫敦的DeepMind公司發佈了該公司深度學習神經網絡AlphaFold 2的一個開源版本，並在《自然》期刊的一篇論文中描述了其方法。該網絡在去年的蛋白質結構預測競賽中"},{"type":"link","attrs":{"href":"https:\/\/www.nature.com\/articles\/d41586-020-03348-4","title":"","type":null},"content":[{"type":"text","text":"取得了領先地位"}]},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/52\/0b\/52bb6d94450e381df54d1d3de272d60b.jpg","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"機器學習軟件預測的人類白細胞介素12蛋白與其受體結合的結構。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"與此同時，一個學術團隊受AlphaFold 2的啓發開發了自己的蛋白質預測工具，該工具已經受到了很多科學家的歡迎。他們的系統稱爲RoseTTaFold，其性能接近AlphaFold 2，具體信息發佈在7月15日發表的《科學》期刊的一篇論文中。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"伊利諾伊州芝加哥大學的計算生物學家Jinbo Xu（他沒有參與這兩個項目）說，這些工具的開源性質意味着科學界應該能夠在前沿技術的基礎上開發出更強大、更有用的軟件。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"從結構到功能"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"蛋白質由氨基酸串組成，摺疊成3D形狀的氨基酸決定了這些蛋白質在細胞中的功能。幾十年來，研究人員一直使用X射線晶體學和冷凍電子顯微鏡等實驗技術來確定蛋白質結構。但是這樣的方法既費時又費錢，而且一些蛋白質不適合這樣的分析。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"去年，DeepMind的突破震動了科學界，這家公司展示了自己的軟件僅使用蛋白質的序列（由DNA決定）就能準確預測許多蛋白質的結構。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"幾十年來，研究人員一直在努力應對這一挑戰，而AlphaFold 2在兩年一度的CASP蛋白質預測競賽中表現如此出色，以至於該競賽的聯合創始人宣稱“從某種意義上說，這個問題已經得到了解決”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"DeepMind（他們以對自身的工作守口如瓶而聞名）12月1日在CASP的一場簡短演講中介紹了AlphaFold 2。它承諾發表一篇更詳細地介紹網絡的論文，並將軟件對研究人員開放，但除此之外就沒多說什麼了。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“在學術界有相當多的悲觀情緒，”西雅圖華盛頓大學的生物化學家David Baker說，他的團隊開發了RoseTTaFold。“如果有人解決了你正在攻關的問題，但沒有透露他們是如何做到的，你接下來該怎麼做呢？”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“當時我感覺自己丟了工作，”Baker團隊的成員、計算化學家Minkyung Baek說。但DeepMind的演講也激發了Baek的很多新想法，讓她迫不及待想要探索一番。因此，她、Baker和他們的同事開始集思廣益，設法複製AlphaFold 2的成功。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"他們確定了幾項關鍵進展，包括網絡如何使用與研究人員試圖預測的目標在進化上相關的蛋白質信息，以及蛋白質某一部分的預測結構如何影響網絡處理與分子其他部分相對應的序列的方式。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"RoseTTaFold不僅表現非常接近AlphaFold 2，而且比其他CASP對手（包括來自Baker實驗室的一些項目）要好得多。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Baek說，目前尚不清楚爲什麼它還比不上AlphaFold 2，但一種可能性來自DeepMind的專業知識。“我們的實驗室裏沒有任何深度學習工程師。”Xu對Baek、Baker和他們同事的努力印象深刻，並猜想DeepMind的成功應該歸功於他們的工程專業知識和卓越的計算能力。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"快速推算結構"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"DeepMind還簡化了AlphaFold 2。AlphaFold首席研究員JohnJumper說，之前該網絡需要幾天的計算時間來爲CASP的某些條目生成結構，但開源版本的速度大約快了16倍。它可以在幾分鐘到幾小時內生成結構，具體取決於蛋白質的大小。這與RoseTTaFold的速度相當。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"儘管AlphaFold 2的源代碼是免費提供的——包括對商業實體也是如此——但對於沒有技術專業知識的研究人員來說，它可能還不是特別有用。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"DeepMind的人工智能科學領域負責人Pushmeet Kohli表示，DeepMind已經在和部分研究人員和組織，包括總部位於瑞士日內瓦的非營利性“被忽視疾病藥物計劃”展開合作，對特定目標進行預測。但這家公司也希望擴大這項技術的應用範圍。“在這個領域，我們還有很多事情要做。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"除了免費提供RoseTTaFold的代碼外，Baker的團隊還建立了一個服務器，研究人員可以在其中插入蛋白質序列並獲得預測的結構。Baker說，自上個月啓動以來，該服務器已經預測了大約500人提交的5,000多種蛋白質的結構。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"現在，RoseTTaFold和AlphaFold 2的代碼都可以免費獲取，研究人員將能夠在這兩項突破的基礎上再接再厲。Xu說，也許科學家可以讓這些技術應對AlphaFold 2之前還難以預測的蛋白質結構。兩個非常受關注的領域分別是預測多種相互作用蛋白質複合物的結構，和將軟件應用於新蛋白質的設計。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"原文鏈接："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/www.nature.com\/articles\/d41586-021-01968-y","title":"","type":null},"content":[{"type":"text","text":"https:\/\/www.nature.com\/articles\/d41586-021-01968-y"}]}]}]}