[AI·洞察] AlphaFold 3 比 AlphaGo 更令人驚嘆 ···"解開生命之謎"

AI，解開生命的秘密··· 改變新藥開發的格局
AI Unlocks the Secrets of Life... Transforming Drug Development

撰文 | 崔奉赫 專欄作家 | The ESG News
Article by Columnist Choi Bong-hyuk | theesg News

蛋白質的結構是理解生命現象的關鍵。 但預測這種結構的工作，數十年來一直是生命科學界的難題。 人工智慧 (AI) 為這個問題提供了答案。 谷歌 DeepMind 開發的 AI 'AlphaFold' 就是主角。
Protein structure is key to understanding life's phenomena. However, predicting this structure remained a decades-long challenge in the life sciences. Artificial intelligence (AI) has provided the answer to this problem. The protagonist is 'AlphaFold,' an AI developed by Google DeepMind.

AlphaFold 是一個 AI 程式，僅使用胺基酸序列就能準確預測蛋白質的三維結構。 簡單來說，AI 首先勾勒出蛋白質的扭曲和轉彎，就像糾結的線一樣長而複雜。 由於蛋白質必須摺疊成特定的結構才能正常發揮功能，AlphaFold 的預測為理解生命現象和開發新藥提供了革命性的見解。
AlphaFold is an AI program that accurately predicts the three-dimensional structure of proteins using only amino acid sequences. Simply put, it means the AI first sketches out the twists and turns of proteins, which are as long and complex as tangled threads. Since proteins must fold into specific structures to function properly, AlphaFold's predictions provide revolutionary insights for understanding life phenomena and developing new drugs.

挑戰「蛋白質摺疊」難題
Challenging the 'Protein Folding' Puzzle

2010 年代中期，DeepMind 以其國際象棋和圍棋 AI 'AlphaGo' 震驚世界後，又迎接了另一個挑戰。 這次是生命科學界通常稱為「聖杯」的『蛋白質摺疊問題』。 過去，揭示蛋白質結構需要數年的實驗和大量的研究資金。 DeepMind 認為，AI 的模式識別能力和龐大的數據資源可以解決這個問題。
In the mid-2010s, after astonishing the world with its chess and Go AI 'AlphaGo,' DeepMind embarked on another challenge. This time, it was the 'protein folding problem,' often called the 'holy grail' of life sciences. Previously, revealing protein structures required years of experiments and enormous research funding. DeepMind believed that AI's pattern recognition capabilities and vast data resources could solve this problem.

在 AI 蛋白質結構預測奧運會上證明了自己的實力
Proving Its Prowess at the AI Protein Structure Prediction Olympics

AlphaFold 於 2018 年首次參加國際蛋白質結構預測競賽 'CASP'。 儘管是 1.0 版本，但它達到了比任何現有預測方法更高的準確度，並獲得了第一名。 當時，它還沒有達到取代實驗的程度，但它證明了 AI 改變生命科學範式的潛力。
AlphaFold made its debut at the 2018 international protein structure prediction competition 'CASP.' Despite being version 1.0, it achieved higher accuracy than any existing prediction method, securing first place. At the time, it wasn't yet at a level to replace experiments, but it proved the potential for AI to change the paradigm of life sciences.

2020 年，AlphaFold 2.0 在同一場比賽中展示了全新的性能水準。 它達到了與實際實驗結果相當甚至超越實際實驗結果的預測準確性。 科學界將此譽為「蛋白質摺疊問題的虛擬解決方案」
In 2020, AlphaFold 2.0 showcased a completely new level of performance at the same competition. It achieved prediction accuracy comparable to or even surpassing actual experimental results. The scientific community hailed this as "the virtual solution to the protein folding problem."

透過資料公開改變科學格局
Changing the Landscape of Science Through Data Disclosure

2021 年，DeepMind 向全球科學家免費提供了 AlphaFold 的預測模型和原始程式碼。 同時，它預測並資料庫化了數百萬個蛋白質的三維結構。 這些數據現在被廣泛用於包括藥物開發、疾病研究和酶工程在內的眾多生命科學領域。 過去需要數年的研究，現在可以在幾分鐘內完成，大大縮短了藥物開發的時間和成本。
In 2021, DeepMind made AlphaFold's prediction model and source code freely available to scientists worldwide. Simultaneously, it predicted and database-ized the 3D structures of millions of proteins. This data is now widely used across numerous life science fields, including drug development, disease research, and enzyme engineering. Research that once took years can now be completed in just minutes, significantly reducing the time and cost of drug development.

https://www.esgre100.com/news/articleView.html?idxno=3586