[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-google-deepmind--alphafold3":3,"tool-google-deepmind--alphafold3":65},[4,23,32,40,49,57],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":22},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",85267,2,"2026-04-18T11:00:28",[13,14,15,16,17,18,19,20,21],"图像","数据工具","视频","插件","Agent","其他","语言模型","开发框架","音频","ready",{"id":24,"name":25,"github_repo":26,"description_zh":27,"stars":28,"difficulty_score":29,"last_commit_at":30,"category_tags":31,"status":22},5784,"funNLP","fighting41love\u002FfunNLP","funNLP 是一个专为中文自然语言处理（NLP）打造的超级资源库，被誉为\"NLP 民工的乐园”。它并非单一的软件工具，而是一个汇集了海量开源项目、数据集、预训练模型和实用代码的综合性平台。\n\n面对中文 NLP 领域资源分散、入门门槛高以及特定场景数据匮乏的痛点，funNLP 提供了“一站式”解决方案。这里不仅涵盖了分词、命名实体识别、情感分析、文本摘要等基础任务的标准工具，还独特地收录了丰富的垂直领域资源，如法律、医疗、金融行业的专用词库与数据集，甚至包含古诗词生成、歌词创作等趣味应用。其核心亮点在于极高的全面性与实用性，从基础的字典词典到前沿的 BERT、GPT-2 模型代码，再到高质量的标注数据和竞赛方案，应有尽有。\n\n无论是刚刚踏入 NLP 领域的学生、需要快速验证想法的算法工程师，还是从事人工智能研究的学者，都能在这里找到急需的“武器弹药”。对于开发者而言，它能大幅减少寻找数据和复现模型的时间；对于研究者，它提供了丰富的基准测试资源和前沿技术参考。funNLP 以开放共享的精神，极大地降低了中文自然语言处理的开发与研究成本，是中文 AI 社区不可或缺的宝藏仓库。",79857,1,"2026-04-08T20:11:31",[19,14,18],{"id":33,"name":34,"github_repo":35,"description_zh":36,"stars":37,"difficulty_score":29,"last_commit_at":38,"category_tags":39,"status":22},5773,"cs-video-courses","Developer-Y\u002Fcs-video-courses","cs-video-courses 是一个精心整理的计算机科学视频课程清单，旨在为自学者提供系统化的学习路径。它汇集了全球知名高校（如加州大学伯克利分校、新南威尔士大学等）的完整课程录像，涵盖从编程基础、数据结构与算法，到操作系统、分布式系统、数据库等核心领域，并深入延伸至人工智能、机器学习、量子计算及区块链等前沿方向。\n\n面对网络上零散且质量参差不齐的教学资源，cs-video-courses 解决了学习者难以找到成体系、高难度大学级别课程的痛点。该项目严格筛选内容，仅收录真正的大学层级课程，排除了碎片化的简短教程或商业广告，确保用户能接触到严谨的学术内容。\n\n这份清单特别适合希望夯实计算机基础的开发者、需要补充特定领域知识的研究人员，以及渴望像在校生一样系统学习计算机科学的自学者。其独特的技术亮点在于分类极其详尽，不仅包含传统的软件工程与网络安全，还细分了生成式 AI、大语言模型、计算生物学等新兴学科，并直接链接至官方视频播放列表，让用户能一站式获取高质量的教育资源，免费享受世界顶尖大学的课堂体验。",79792,"2026-04-08T22:03:59",[18,13,14,20],{"id":41,"name":42,"github_repo":43,"description_zh":44,"stars":45,"difficulty_score":46,"last_commit_at":47,"category_tags":48,"status":22},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,3,"2026-04-04T04:44:48",[17,13,20,19,18],{"id":50,"name":51,"github_repo":52,"description_zh":53,"stars":54,"difficulty_score":46,"last_commit_at":55,"category_tags":56,"status":22},519,"PaddleOCR","PaddlePaddle\u002FPaddleOCR","PaddleOCR 是一款基于百度飞桨框架开发的高性能开源光学字符识别工具包。它的核心能力是将图片、PDF 等文档中的文字提取出来，转换成计算机可读取的结构化数据，让机器真正“看懂”图文内容。\n\n面对海量纸质或电子文档，PaddleOCR 解决了人工录入效率低、数字化成本高的问题。尤其在人工智能领域，它扮演着连接图像与大型语言模型（LLM）的桥梁角色，能将视觉信息直接转化为文本输入，助力智能问答、文档分析等应用场景落地。\n\nPaddleOCR 适合开发者、算法研究人员以及有文档自动化需求的普通用户。其技术优势十分明显：不仅支持全球 100 多种语言的识别，还能在 Windows、Linux、macOS 等多个系统上运行，并灵活适配 CPU、GPU、NPU 等各类硬件。作为一个轻量级且社区活跃的开源项目，PaddleOCR 既能满足快速集成的需求，也能支撑前沿的视觉语言研究，是处理文字识别任务的理想选择。",75872,"2026-04-18T10:54:57",[19,13,20,18],{"id":58,"name":59,"github_repo":60,"description_zh":61,"stars":62,"difficulty_score":29,"last_commit_at":63,"category_tags":64,"status":22},3215,"awesome-machine-learning","josephmisiti\u002Fawesome-machine-learning","awesome-machine-learning 是一份精心整理的机器学习资源清单，汇集了全球优秀的机器学习框架、库和软件工具。面对机器学习领域技术迭代快、资源分散且难以甄选的痛点，这份清单按编程语言（如 Python、C++、Go 等）和应用场景（如计算机视觉、自然语言处理、深度学习等）进行了系统化分类，帮助使用者快速定位高质量项目。\n\n它特别适合开发者、数据科学家及研究人员使用。无论是初学者寻找入门库，还是资深工程师对比不同语言的技术选型，都能从中获得极具价值的参考。此外，清单还延伸提供了免费书籍、在线课程、行业会议、技术博客及线下聚会等丰富资源，构建了从学习到实践的全链路支持体系。\n\n其独特亮点在于严格的维护标准：明确标记已停止维护或长期未更新的项目，确保推荐内容的时效性与可靠性。作为机器学习领域的“导航图”，awesome-machine-learning 以开源协作的方式持续更新，旨在降低技术探索门槛，让每一位从业者都能高效地站在巨人的肩膀上创新。",72149,"2026-04-03T21:50:24",[20,18],{"id":66,"github_repo":67,"name":68,"description_en":69,"description_zh":70,"ai_summary_zh":70,"readme_en":71,"readme_zh":72,"quickstart_zh":73,"use_case_zh":74,"hero_image_url":75,"owner_login":76,"owner_name":77,"owner_avatar_url":78,"owner_bio":79,"owner_company":80,"owner_location":80,"owner_email":80,"owner_twitter":80,"owner_website":81,"owner_url":82,"languages":83,"stars":104,"forks":105,"last_commit_at":106,"license":107,"difficulty_score":108,"env_os":109,"env_gpu":110,"env_ram":111,"env_deps":112,"category_tags":125,"github_topics":80,"view_count":10,"oss_zip_url":80,"oss_zip_packed_at":80,"status":22,"created_at":126,"updated_at":127,"faqs":128,"releases":158},9200,"google-deepmind\u002Falphafold3","alphafold3","AlphaFold 3 inference pipeline.","AlphaFold 3 是由 Google DeepMind 推出的前沿生物分子结构预测工具，旨在精准模拟蛋白质、核酸、配体及修饰物之间的复杂相互作用。它解决了传统方法难以高效解析多组分生物复合物三维结构的难题，将原本耗时数月的实验过程缩短至分钟级，极大加速了新药研发与基础生命科学探索。\n\n这款工具主要面向生物信息学研究人员、计算生物学家以及具备一定编程基础的开发者。用户需通过官方申请获取模型参数，并依托本地 GPU 环境运行推理流程。相比前代产品，AlphaFold 3 的核心突破在于其统一的架构设计，不再局限于单一蛋白质折叠，而是能同时处理包含多种生物分子的综合体系，显著提升了预测的广度与精度。虽然官方也提供了受限的网页版服务，但开源版本允许科研人员在合规前提下深度定制数据流水线，进行更灵活的学术研究与验证。对于渴望在原子层面理解生命机制的团队而言，AlphaFold 3 无疑是一把开启微观世界大门的关键钥匙。","![header](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_alphafold3_readme_d9339130ec93.jpg)\n\n# AlphaFold 3\n\nThis package provides an implementation of the inference pipeline of AlphaFold\n3. See below for how to access the model parameters. You may only use AlphaFold\n3 model parameters if received directly from Google. Use is subject to these\n[terms of use](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md).\n\nAny publication that discloses findings arising from using this source code, the\nmodel parameters or outputs produced by those should [cite](#citing-this-work)\nthe\n[Accurate structure prediction of biomolecular interactions with AlphaFold 3](https:\u002F\u002Fdoi.org\u002F10.1038\u002Fs41586-024-07487-w)\npaper.\n\nPlease also refer to the Supplementary Information for a detailed description of\nthe method.\n\nAlphaFold 3 is also available at\n[alphafoldserver.com](https:\u002F\u002Falphafoldserver.com) for non-commercial use,\nthough with a more limited set of ligands and covalent modifications.\n\nIf you have any questions, please contact the AlphaFold team at\n[alphafold@google.com](mailto:alphafold@google.com).\n\n## Obtaining Model Parameters\n\nThis repository contains all necessary code for AlphaFold 3 inference. To\nrequest access to the AlphaFold 3 model parameters, please complete\n[this form](https:\u002F\u002Fforms.gle\u002FsvvpY4u2jsHEwWYS6). Access will be granted at\nGoogle DeepMind’s sole discretion. We will aim to respond to requests within 2–3\nbusiness days. You may only use AlphaFold 3 model parameters if received\ndirectly from Google. Use is subject to these\n[terms of use](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md).\n\n## Installation and Running Your First Prediction\n\nSee the [installation documentation](docs\u002Finstallation.md).\n\nOnce you have installed AlphaFold 3, you can test your setup using e.g. the\nfollowing input JSON file named `fold_input.json`:\n\n```json\n{\n  \"name\": \"2PV7\",\n  \"sequences\": [\n    {\n      \"protein\": {\n        \"id\": [\"A\", \"B\"],\n        \"sequence\": \"GMRESYANENQFGFKTINSDIHKIVIVGGYGKLGGLFARYLRASGYPISILDREDWAVAESILANADVVIVSVPINLTLETIERLKPYLTENMLLADLTSVKREPLAKMLEVHTGAVLGLHPMFGADIASMAKQVVVRCDGRFPERYEWLLEQIQIWGAKIYQTNATEHDHNMTYIQALRHFSTFANGLHLSKQPINLANLLALSSPIYRLELAMIGRLFAQDAELYADIIMDKSENLAVIETLKQTYDEALTFFENNDRQGFIDAFHKVRDWFGDYSEQFLKESRQLLQQANDLKQG\"\n      }\n    }\n  ],\n  \"modelSeeds\": [1],\n  \"dialect\": \"alphafold3\",\n  \"version\": 1\n}\n```\n\nYou can then run AlphaFold 3 using the following command:\n\n```\ndocker run -it \\\n    --volume $HOME\u002Faf_input:\u002Froot\u002Faf_input \\\n    --volume $HOME\u002Faf_output:\u002Froot\u002Faf_output \\\n    --volume \u003CMODEL_PARAMETERS_DIR>:\u002Froot\u002Fmodels \\\n    --volume \u003CDATABASES_DIR>:\u002Froot\u002Fpublic_databases \\\n    --gpus all \\\n    alphafold3 \\\n    python run_alphafold.py \\\n    --json_path=\u002Froot\u002Faf_input\u002Ffold_input.json \\\n    --model_dir=\u002Froot\u002Fmodels \\\n    --output_dir=\u002Froot\u002Faf_output\n```\n\nThere are various flags that you can pass to the `run_alphafold.py` command, to\nlist them all run `python run_alphafold.py --help`. Two fundamental flags that\ncontrol which parts AlphaFold 3 will run are:\n\n*   `--run_data_pipeline` (defaults to `true`): whether to run the data\n    pipeline, i.e. genetic and template search. This part is CPU-only, time\n    consuming and could be run on a machine without a GPU.\n*   `--run_inference` (defaults to `true`): whether to run the inference. This\n    part requires a GPU.\n\n## AlphaFold 3 Input\n\nSee the [input documentation](docs\u002Finput.md).\n\n## AlphaFold 3 Output\n\nSee the [output documentation](docs\u002Foutput.md).\n\n## Performance\n\nSee the [performance documentation](docs\u002Fperformance.md).\n\n## Known Issues\n\nKnown issues are documented in the\n[known issues documentation](docs\u002Fknown_issues.md).\n\nPlease\n[create an issue](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fissues\u002Fnew\u002Fchoose)\nif it is not already listed in [Known Issues](docs\u002Fknown_issues.md) or in the\n[issues tracker](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fissues).\n\n## Citing This Work\n\nAny publication that discloses findings arising from using this source code, the\nmodel parameters or outputs produced by those should cite:\n\n```bibtex\n@article{Abramson2024,\n  author  = {Abramson, Josh and Adler, Jonas and Dunger, Jack and Evans, Richard and Green, Tim and Pritzel, Alexander and Ronneberger, Olaf and Willmore, Lindsay and Ballard, Andrew J. and Bambrick, Joshua and Bodenstein, Sebastian W. and Evans, David A. and Hung, Chia-Chun and O’Neill, Michael and Reiman, David and Tunyasuvunakool, Kathryn and Wu, Zachary and Žemgulytė, Akvilė and Arvaniti, Eirini and Beattie, Charles and Bertolli, Ottavia and Bridgland, Alex and Cherepanov, Alexey and Congreve, Miles and Cowen-Rivers, Alexander I. and Cowie, Andrew and Figurnov, Michael and Fuchs, Fabian B. and Gladman, Hannah and Jain, Rishub and Khan, Yousuf A. and Low, Caroline M. R. and Perlin, Kuba and Potapenko, Anna and Savy, Pascal and Singh, Sukhdeep and Stecula, Adrian and Thillaisundaram, Ashok and Tong, Catherine and Yakneen, Sergei and Zhong, Ellen D. and Zielinski, Michal and Žídek, Augustin and Bapst, Victor and Kohli, Pushmeet and Jaderberg, Max and Hassabis, Demis and Jumper, John M.},\n  journal = {Nature},\n  title   = {Accurate structure prediction of biomolecular interactions with AlphaFold 3},\n  year    = {2024},\n  volume  = {630},\n  number  = {8016},\n  pages   = {493–-500},\n  doi     = {10.1038\u002Fs41586-024-07487-w}\n}\n```\n\n## Acknowledgements\n\nAlphaFold 3's release was made possible by the invaluable contributions of the\nfollowing people:\n\nAndrew Cowie, Bella Hansen, Charlie Beattie, Chris Jones, Grace Margand,\nJacob Kelly, James Spencer, Josh Abramson, Kathryn Tunyasuvunakool, Kuba Perlin,\nLindsay Willmore, Max Bileschi, Molly Beck, Oleg Kovalevskiy,\nSebastian Bodenstein, Sukhdeep Singh, Tim Green, Toby Sargeant, Uchechi Okereke,\nYotam Doron, and Augustin Žídek (engineering lead).\n\nWe also extend our gratitude to our collaborators at Google and Isomorphic Labs.\n\nAlphaFold 3 uses the following separate libraries and packages:\n\n*   [abseil-cpp](https:\u002F\u002Fgithub.com\u002Fabseil\u002Fabseil-cpp) and\n    [abseil-py](https:\u002F\u002Fgithub.com\u002Fabseil\u002Fabseil-py)\n*   [Docker](https:\u002F\u002Fwww.docker.com)\n*   [DSSP](https:\u002F\u002Fgithub.com\u002FPDB-REDO\u002Fdssp)\n*   [HMMER Suite](https:\u002F\u002Fgithub.com\u002FEddyRivasLab\u002Fhmmer)\n*   [Haiku](https:\u002F\u002Fgithub.com\u002Fdeepmind\u002Fdm-haiku)\n*   [JAX](https:\u002F\u002Fgithub.com\u002Fjax-ml\u002Fjax\u002F)\n*   [libcifpp](https:\u002F\u002Fgithub.com\u002Fpdb-redo\u002Flibcifpp)\n*   [NumPy](https:\u002F\u002Fgithub.com\u002Fnumpy\u002Fnumpy)\n*   [pybind11](https:\u002F\u002Fgithub.com\u002Fpybind\u002Fpybind11) and\n    [pybind11_abseil](https:\u002F\u002Fgithub.com\u002Fpybind\u002Fpybind11_abseil)\n*   [RDKit](https:\u002F\u002Fgithub.com\u002Frdkit\u002Frdkit)\n*   [Tokamax](https:\u002F\u002Fgithub.com\u002Fopenxla\u002Ftokamax)\n*   [tqdm](https:\u002F\u002Fgithub.com\u002Ftqdm\u002Ftqdm)\n\nWe thank all their contributors and maintainers!\n\n## Get in Touch\n\nIf you have any questions not covered in this overview, please contact the\nAlphaFold team at alphafold@google.com.\n\nWe would love to hear your feedback and understand how AlphaFold 3 has been\nuseful in your research. Share your stories with us at\n[alphafold@google.com](mailto:alphafold@google.com).\n\n## Licence and Disclaimer\n\nThis is not an officially supported Google product.\n\nCopyright 2024 DeepMind Technologies Limited.\n\n### AlphaFold 3 Source Code and Model Parameters\n\nThe AlphaFold 3 source code is licensed under the Creative Commons\nAttribution-Non-Commercial ShareAlike International License, Version 4.0\n(CC-BY-NC-SA 4.0) (the \"License\"); you may not use this file except in\ncompliance with the License. You may obtain a copy of the License at\n[https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FLICENSE](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FLICENSE).\n\nThe AlphaFold 3 model parameters are made available under the\n[AlphaFold 3 Model Parameters Terms of Use](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md)\n(the \"Terms\"); you may not use these except in compliance with the Terms. You\nmay obtain a copy of the Terms at\n[https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md).\n\nUnless required by applicable law, AlphaFold 3 and its output are distributed on\nan \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express\nor implied. You are solely responsible for determining the appropriateness of\nusing AlphaFold 3, or using or distributing its source code or output, and\nassume any and all risks associated with such use or distribution and your\nexercise of rights and obligations under the relevant terms. Output are\npredictions with varying levels of confidence and should be interpreted\ncarefully. Use discretion before relying on, publishing, downloading or\notherwise using the AlphaFold 3 Assets.\n\nAlphaFold 3 and its output are for theoretical modeling only. They are not\nintended, validated, or approved for clinical use. You should not use the\nAlphaFold 3 or its output for clinical purposes or rely on them for medical or\nother professional advice. Any content regarding those topics is provided for\ninformational purposes only and is not a substitute for advice from a qualified\nprofessional. See the relevant terms for the specific language governing\npermissions and limitations under the terms.\n\n### Third-party Software\n\nUse of the third-party software, libraries or code referred to in the\n[Acknowledgements](#acknowledgements) section above may be governed by separate\nterms and conditions or license provisions. Your use of the third-party\nsoftware, libraries or code is subject to any such terms and you should check\nthat you can comply with any applicable restrictions or terms and conditions\nbefore use.\n\n### Mirrored and Reference Databases\n\nThe following databases have been: (1) mirrored by Google DeepMind; and (2) in\npart, included with the inference code package for testing purposes, and are\navailable with reference to the following:\n\n*   [BFD](https:\u002F\u002Fbfd.mmseqs.com\u002F) (modified), by Steinegger M. and Söding J.,\n    modified by Google DeepMind, available under a\n    [Creative Commons Attribution 4.0 International License](https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002Fdeed.en).\n    See the Methods section of the\n    [AlphaFold proteome paper](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-021-03828-1)\n    for details.\n*   [PDB](https:\u002F\u002Fwwpdb.org) (unmodified), by H.M. Berman et al., available free\n    of all copyright restrictions and made fully and freely available for both\n    non-commercial and commercial use under\n    [CC0 1.0 Universal (CC0 1.0) Public Domain Dedication](https:\u002F\u002Fcreativecommons.org\u002Fpublicdomain\u002Fzero\u002F1.0\u002F).\n*   [MGnify: v2022\\_05](https:\u002F\u002Fftp.ebi.ac.uk\u002Fpub\u002Fdatabases\u002Fmetagenomics\u002Fpeptide_database\u002F2022_05\u002FREADME.txt)\n    (unmodified), by Mitchell AL et al., available free of all copyright\n    restrictions and made fully and freely available for both non-commercial and\n    commercial use under\n    [CC0 1.0 Universal (CC0 1.0) Public Domain Dedication](https:\u002F\u002Fcreativecommons.org\u002Fpublicdomain\u002Fzero\u002F1.0\u002F).\n*   [UniProt: 2021\\_04](https:\u002F\u002Fwww.uniprot.org\u002F) (unmodified), by The UniProt\n    Consortium, available under a\n    [Creative Commons Attribution 4.0 International License](https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002Fdeed.en).\n*   [UniRef90: 2022\\_05](https:\u002F\u002Fwww.uniprot.org\u002F) (unmodified) by The UniProt\n    Consortium, available under a\n    [Creative Commons Attribution 4.0 International License](https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002Fdeed.en).\n*   [NT: 2023\\_02\\_23](https:\u002F\u002Fwww.ncbi.nlm.nih.gov\u002Fnucleotide\u002F) (modified) See\n    the Supplementary Information of the\n    [AlphaFold 3 paper](https:\u002F\u002Fnature.com\u002Farticles\u002Fs41586-024-07487-w) for\n    details.\n*   [RFam: 14\\_4](https:\u002F\u002Frfam.org\u002F) (modified), by I. Kalvari et al., available\n    free of all copyright restrictions and made fully and freely available for\n    both non-commercial and commercial use under\n    [CC0 1.0 Universal (CC0 1.0) Public Domain Dedication](https:\u002F\u002Fcreativecommons.org\u002Fpublicdomain\u002Fzero\u002F1.0\u002F).\n    See the Supplementary Information of the\n    [AlphaFold 3 paper](https:\u002F\u002Fnature.com\u002Farticles\u002Fs41586-024-07487-w) for\n    details.\n*   [RNACentral: 21\\_0](https:\u002F\u002Frnacentral.org\u002F) (modified), by The RNAcentral\n    Consortium available free of all copyright restrictions and made fully and\n    freely available for both non-commercial and commercial use under\n    [CC0 1.0 Universal (CC0 1.0) Public Domain Dedication](https:\u002F\u002Fcreativecommons.org\u002Fpublicdomain\u002Fzero\u002F1.0\u002F).\n    See the Supplementary Information of the\n    [AlphaFold 3 paper](https:\u002F\u002Fnature.com\u002Farticles\u002Fs41586-024-07487-w) for\n    details.\n","![header](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_alphafold3_readme_d9339130ec93.jpg)\n\n# AlphaFold 3\n\n本软件包提供了 AlphaFold 3 推理流程的实现。有关如何获取模型参数的信息，请参见下文。您仅可在直接从 Google 获得的情况下使用 AlphaFold 3 模型参数。使用须遵守这些[使用条款](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md)。\n\n任何披露基于本源代码、模型参数或其生成输出的研究成果的出版物，均应[引用](#citing-this-work)论文《利用 AlphaFold 3 准确预测生物分子相互作用的结构》（DOI: 10.1038\u002Fs41586-024-07487-w）。\n\n此外，请参阅补充信息部分，以获取方法的详细描述。\n\nAlphaFold 3 亦可通过 [alphafoldserver.com](https:\u002F\u002Falphafoldserver.com) 提供非商业用途的服务，但其配体和共价修饰的种类较为有限。\n\n如有任何疑问，请联系 AlphaFold 团队：[alphafold@google.com](mailto:alphafold@google.com)。\n\n## 获取模型参数\n\n本仓库包含运行 AlphaFold 3 推理所需的所有代码。如需申请访问 AlphaFold 3 模型参数，请填写[此表格](https:\u002F\u002Fforms.gle\u002FsvvpY4u2jsHEwWYS6)。访问权限将由 Google DeepMind 全权决定。我们力争在 2–3 个工作日内回复您的请求。您仅可在直接从 Google 获得的情况下使用 AlphaFold 3 模型参数。使用须遵守这些[使用条款](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md)。\n\n## 安装与首次预测运行\n\n请参阅[安装文档](docs\u002Finstallation.md)。\n\n完成 AlphaFold 3 的安装后，您可以使用以下名为 `fold_input.json` 的输入 JSON 文件来测试您的设置：\n\n```json\n{\n  \"name\": \"2PV7\",\n  \"sequences\": [\n    {\n      \"protein\": {\n        \"id\": [\"A\", \"B\"],\n        \"sequence\": \"GMRESYANENQFGFKTINSDIHKIVIVGGYGKLGGLFARYLRASGYPISILDREDWAVAESILANADVVIVSVPINLTLETIERLKPYLTENMLLADLTSVKREPLAKMLEVHTGAVLGLHPMFGADIASMAKQVVVRCDGRFPERYEWLLEQIQIWGAKIYQTNATEHDHNMTYIQALRHFSTFANGLHLSKQPINLANLLALSSPIYRLELAMIGRLFAQDAELYADIIMDKSENLAVIETLKQTYDEALTFFENNDRQGFIDAFHKVRDWFGDYSEQFLKESRQLLQQANDLKQG\"\n      }\n    }\n  ],\n  \"modelSeeds\": [1],\n  \"dialect\": \"alphafold3\",\n  \"version\": 1\n}\n```\n\n随后，您可使用以下命令运行 AlphaFold 3：\n\n```\ndocker run -it \\\n    --volume $HOME\u002Faf_input:\u002Froot\u002Faf_input \\\n    --volume $HOME\u002Faf_output:\u002Froot\u002Faf_output \\\n    --volume \u003CMODEL_PARAMETERS_DIR>:\u002Froot\u002Fmodels \\\n    --volume \u003CDATABASES_DIR>:\u002Froot\u002Fpublic_databases \\\n    --gpus all \\\n    alphafold3 \\\n    python run_alphafold.py \\\n    --json_path=\u002Froot\u002Faf_input\u002Ffold_input.json \\\n    --model_dir=\u002Froot\u002Fmodels \\\n    --output_dir=\u002Froot\u002Faf_output\n```\n\n`run_alphafold.py` 命令支持多种参数选项，您可以通过运行 `python run_alphafold.py --help` 查看所有可用选项。其中两个关键参数用于控制 AlphaFold 3 将执行哪些步骤：\n\n*   `--run_data_pipeline`（默认为 `true`）：是否运行数据处理流程，即基因组和模板搜索。该步骤仅需 CPU，耗时较长，也可在无 GPU 的机器上运行。\n*   `--run_inference`（默认为 `true`）：是否运行推理过程。此步骤需要 GPU。\n\n## AlphaFold 3 输入\n\n请参阅[输入文档](docs\u002Finput.md)。\n\n## AlphaFold 3 输出\n\n请参阅[输出文档](docs\u002Foutput.md)。\n\n## 性能\n\n请参阅[性能文档](docs\u002Fperformance.md)。\n\n## 已知问题\n\n已知问题已在[已知问题文档](docs\u002Fknown_issues.md)中列出。\n\n如果问题未在[已知问题](docs\u002Fknown_issues.md)或[问题追踪器](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fissues)中列出，请[创建新问题](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fissues\u002Fnew\u002Fchoose)。\n\n## 引用本工作\n\n任何披露基于本源代码、模型参数或其生成输出的研究成果的出版物，均应引用以下文献：\n\n```bibtex\n@article{Abramson2024,\n  author  = {Abramson, Josh and Adler, Jonas and Dunger, Jack and Evans, Richard and Green, Tim and Pritzel, Alexander and Ronneberger, Olaf and Willmore, Lindsay and Ballard, Andrew J. and Bambrick, Joshua and Bodenstein, Sebastian W. and Evans, David A. and Hung, Chia-Chun and O’Neill, Michael and Reiman, David and Tunyasuvunakool, Kathryn and Wu, Zachary and Žemgulytė, Akvilė and Arvaniti, Eirini and Beattie, Charles and Bertolli, Ottavia and Bridgland, Alex and Cherepanov, Alexey and Congreve, Miles and Cowen-Rivers, Alexander I. and Cowie, Andrew and Figurnov, Michael and Fuchs, Fabian B. and Gladman, Hannah and Jain, Rishub and Khan, Yousuf A. and Low, Caroline M. R. and Perlin, Kuba and Potapenko, Anna and Savy, Pascal and Singh, Sukhdeep and Stecula, Adrian and Thillaisundaram, Ashok and Tong, Catherine and Yakneen, Sergei and Zhong, Ellen D. and Zielinski, Michal and Žídek, Augustin and Bapst, Victor and Kohli, Pushmeet and Jaderberg, Max and Hassabis, Demis and Jumper, John M.},\n  journal = {Nature},\n  title   = {Accurate structure prediction of biomolecular interactions with AlphaFold 3},\n  year    = {2024},\n  volume  = {630},\n  number  = {8016},\n  pages   = {493–500},\n  doi     = {10.1038\u002Fs41586-024-07487-w}\n}\n```\n\n## 致谢\n\nAlphaFold 3 的发布离不开以下各位的宝贵贡献：\n\nAndrew Cowie、Bella Hansen、Charlie Beattie、Chris Jones、Grace Margand、Jacob Kelly、James Spencer、Josh Abramson、Kathryn Tunyasuvunakool、Kuba Perlin、Lindsay Willmore、Max Bileschi、Molly Beck、Oleg Kovalevskiy、Sebastian Bodenstein、Sukhdeep Singh、Tim Green、Toby Sargeant、Uchechi Okereke、Yotam Doron 以及工程负责人 Augustin Žídek。\n\n我们亦感谢 Google 和 Isomorphic Labs 的合作伙伴们。\n\nAlphaFold 3 使用了以下独立的库和软件包：\n\n*   [abseil-cpp](https:\u002F\u002Fgithub.com\u002Fabseil\u002Fabseil-cpp) 和 [abseil-py](https:\u002F\u002Fgithub.com\u002Fabseil\u002Fabseil-py)\n*   [Docker](https:\u002F\u002Fwww.docker.com)\n*   [DSSP](https:\u002F\u002Fgithub.com\u002FPDB-REDO\u002Fdssp)\n*   [HMMER Suite](https:\u002F\u002Fgithub.com\u002FEddyRivasLab\u002Fhmmer)\n*   [Haiku](https:\u002F\u002Fgithub.com\u002Fdeepmind\u002Fdm-haiku)\n*   [JAX](https:\u002F\u002Fgithub.com\u002Fjax-ml\u002Fjax\u002F)\n*   [libcifpp](https:\u002F\u002Fgithub.com\u002Fpdb-redo\u002Flibcifpp)\n*   [NumPy](https:\u002F\u002Fgithub.com\u002Fnumpy\u002Fnumpy)\n*   [pybind11](https:\u002F\u002Fgithub.com\u002Fpybind\u002Fpybind11) 和 [pybind11_abseil](https:\u002F\u002Fgithub.com\u002Fpybind\u002Fpybind11_abseil)\n*   [RDKit](https:\u002F\u002Fgithub.com\u002Frdkit\u002Frdkit)\n*   [Tokamax](https:\u002F\u002Fgithub.com\u002Fopenxla\u002Ftokamax)\n*   [tqdm](https:\u002F\u002Fgithub.com\u002Ftqdm\u002Ftqdm)\n\n我们衷心感谢所有贡献者和维护者！\n\n## 联系我们\n\n如果您有任何本概述未涵盖的问题，请通过 alphafold@google.com 联系 AlphaFold 团队。\n\n我们非常期待您的反馈，并希望了解 AlphaFold 3 在您的研究中如何发挥作用。请将您的故事分享给我们，发送至 [alphafold@google.com](mailto:alphafold@google.com)。\n\n## 许可与免责声明\n\n本产品并非 Google 官方支持的产品。\n\n版权所有 © 2024 DeepMind Technologies Limited。\n\n### AlphaFold 3 源代码和模型参数\n\nAlphaFold 3 的源代码根据知识共享署名-非商业性使用-相同方式共享 4.0 国际许可协议（CC-BY-NC-SA 4.0）授权；您不得在不符合该许可协议的情况下使用此文件。您可以在 [https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FLICENSE](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FLICENSE) 获取该许可协议的副本。\n\nAlphaFold 3 的模型参数则依据 [AlphaFold 3 模型参数使用条款](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md)（“条款”）提供；您不得在不符合这些条款的情况下使用它们。您可以在 [https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md) 获取该条款的副本。\n\n除非适用法律另有要求，否则 AlphaFold 3 及其输出均按“现状”提供，不附带任何形式的保证或条件，无论是明示的还是默示的。您需自行判断是否适合使用 AlphaFold 3，或使用、分发其源代码或输出，并承担与此类使用或分发以及您在相关条款下行使权利和履行义务有关的一切风险。输出结果为具有不同置信度的预测，应谨慎解读。在依赖、发布、下载或以其他方式使用 AlphaFold 3 相关资源之前，请务必谨慎决策。\n\nAlphaFold 3 及其输出仅用于理论建模，不适用于临床用途，也未经验证或批准用于临床。您不应将 AlphaFold 3 或其输出用于临床目的，亦不应依赖其结果获取医疗或其他专业建议。任何涉及此类主题的内容仅供信息参考，不能替代合格专业人士的建议。具体权限与限制条款请参阅相关使用条款。\n\n### 第三方软件\n\n上文“致谢”部分提及的第三方软件、库或代码可能受单独的条款、条件或许可协议约束。您对这些第三方软件、库或代码的使用须遵守相应规定，在使用前应确认自己能够符合所有适用的限制或条款与条件。\n\n### 镜像及参考数据库\n\n以下数据库由 Google DeepMind 进行了镜像，并部分包含于推理代码包中以供测试之用，现可供参考：\n\n*   [BFD](https:\u002F\u002Fbfd.mmseqs.com\u002F)（已修改），由 Steinegger M. 和 Söding J. 创建，经 Google DeepMind 修改后，依据 [知识共享署名 4.0 国际许可协议](https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002Fdeed.en) 提供。详细信息请参阅 [AlphaFold 蛋白质组论文](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-021-03828-1) 的方法部分。\n*   [PDB](https:\u002F\u002Fwwpdb.org)（未修改），由 H.M. Berman 等人创建，现已完全免除所有版权限制，并依据 [CC0 1.0 全球公共领域奉献协议](https:\u002F\u002Fcreativecommons.org\u002Fpublicdomain\u002Fzero\u002F1.0\u002F) 对非商业及商业用途开放。\n*   [MGnify: v2022_05](https:\u002F\u002Fftp.ebi.ac.uk\u002Fpub\u002Fdatabases\u002Fmetagenomics\u002Fpeptide_database\u002F2022_05\u002FREADME.txt)（未修改），由 Mitchell AL 等人创建，同样免除了所有版权限制，并依据 [CC0 1.0 全球公共领域奉献协议](https:\u002F\u002Fcreativecommons.org\u002Fpublicdomain\u002Fzero\u002F1.0\u002F) 对非商业及商业用途完全开放。\n*   [UniProt: 2021_04](https:\u002F\u002Fwww.uniprot.org\u002F)（未修改），由 UniProt 联盟创建，依据 [知识共享署名 4.0 国际许可协议](https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002Fdeed.en) 提供。\n*   [UniRef90: 2022_05](https:\u002F\u002Fwww.uniprot.org\u002F)（未修改），同样由 UniProt 联盟创建，依据 [知识共享署名 4.0 国际许可协议](https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002Fdeed.en) 提供。\n*   [NT: 2023_02_23](https:\u002F\u002Fwww.ncbi.nlm.nih.gov\u002Fnucleotide\u002F)（已修改）。详情请参阅 [AlphaFold 3 论文补充信息](https:\u002F\u002Fnature.com\u002Farticles\u002Fs41586-024-07487-w)。\n*   [RFam: 14_4](https:\u002F\u002Frfam.org\u002F)（已修改），由 I. Kalvari 等人创建，完全免除了所有版权限制，并依据 [CC0 1.0 全球公共领域奉献协议](https:\u002F\u002Fcreativecommons.org\u002Fpublicdomain\u002Fzero\u002F1.0\u002F) 对非商业及商业用途完全开放。详细信息请参阅 [AlphaFold 3 论文补充信息](https:\u002F\u002Fnature.com\u002Farticles\u002Fs41586-024-07487-w)。\n*   [RNACentral: 21_0](https:\u002F\u002Frnacentral.org\u002F)（已修改），由 RNAcentral 联盟创建，同样免除了所有版权限制，并依据 [CC0 1.0 全球公共领域奉献协议](https:\u002F\u002Fcreativecommons.org\u002Fpublicdomain\u002Fzero\u002F1.0\u002F) 对非商业及商业用途完全开放。详细信息请参阅 [AlphaFold 3 论文补充信息](https:\u002F\u002Fnature.com\u002Farticles\u002Fs41586-024-07487-w)。","# AlphaFold 3 快速上手指南\n\n## 环境准备\n\n### 系统要求\n- **操作系统**: Linux (推荐 Ubuntu 20.04 或更高版本)\n- **GPU**: 支持 CUDA 的 NVIDIA 显卡（推理阶段必需）\n- **Docker**: 已安装并配置好 NVIDIA Container Toolkit (`nvidia-docker`)\n- **磁盘空间**: 建议预留至少 500GB 空间用于存放模型参数和生物数据库\n\n### 前置依赖\n- 确保宿主机已安装 Docker 和 GPU 驱动。\n- **模型参数与数据库**: \n  - 源代码可免费获取，但**模型参数**需向 Google DeepMind 单独申请。请访问 [申请表单](https:\u002F\u002Fforms.gle\u002FsvvpY4u2jsHEwWYS6) 提交申请，审核通过后会将参数直接发送给您。\n  - 使用前请仔细阅读并同意 [模型参数使用条款](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fblob\u002Fmain\u002FWEIGHTS_TERMS_OF_USE.md)。\n  - 您需要自行准备遗传序列数据库（如 BFD, UniProt 等），具体列表参考官方文档。\n\n## 安装步骤\n\nAlphaFold 3 推荐使用 Docker 容器运行，无需在宿主机复杂配置 Python 环境。\n\n1. **拉取 Docker 镜像**\n   构建或拉取官方提供的 AlphaFold 3 镜像（假设您已克隆源码仓库）：\n   ```bash\n   docker build -t alphafold3 .\n   ```\n   *注：如果官方提供预构建镜像，可直接使用 `docker pull` 命令。*\n\n2. **目录结构规划**\n   在宿主机创建必要的挂载目录，用于存放输入、输出、模型参数及数据库：\n   ```bash\n   mkdir -p $HOME\u002Faf_input $HOME\u002Faf_output $HOME\u002Fmodels $HOME\u002Fpublic_databases\n   ```\n   - 将申请到的**模型参数文件**放入 `$HOME\u002Fmodels`。\n   - 将下载的**生物数据库文件**放入 `$HOME\u002Fpublic_databases`。\n\n## 基本使用\n\n### 1. 准备输入文件\n创建一个名为 `fold_input.json` 的文件，定义待预测的蛋白质序列。以下是一个最简单的单蛋白预测示例：\n\n```json\n{\n  \"name\": \"2PV7\",\n  \"sequences\": [\n    {\n      \"protein\": {\n        \"id\": [\"A\"],\n        \"sequence\": \"GMRESYANENQFGFKTINSDIHKIVIVGGYGKLGGLFARYLRASGYPISILDREDWAVAESILANADVVIVSVPINLTLETIERLKPYLTENMLLADLTSVKREPLAKMLEVHTGAVLGLHPMFGADIASMAKQVVVRCDGRFPERYEWLLEQIQIWGAKIYQTNATEHDHNMTYIQALRHFSTFANGLHLSKQPINLANLLALSSPIYRLELAMIGRLFAQDAELYADIIMDKSENLAVIETLKQTYDEALTFFENNDRQGFIDAFHKVRDWFGDYSEQFLKESRQLLQQANDLKQG\"\n      }\n    }\n  ],\n  \"modelSeeds\": [1],\n  \"dialect\": \"alphafold3\",\n  \"version\": 1\n}\n```\n将该文件保存至 `$HOME\u002Faf_input\u002Ffold_input.json`。\n\n### 2. 运行预测\n使用以下 Docker 命令启动预测流程。该命令将自动执行数据检索（如需）和模型推理：\n\n```bash\ndocker run -it \\\n    --volume $HOME\u002Faf_input:\u002Froot\u002Faf_input \\\n    --volume $HOME\u002Faf_output:\u002Froot\u002Faf_output \\\n    --volume $HOME\u002Fmodels:\u002Froot\u002Fmodels \\\n    --volume $HOME\u002Fpublic_databases:\u002Froot\u002Fpublic_databases \\\n    --gpus all \\\n    alphafold3 \\\n    python run_alphafold.py \\\n    --json_path=\u002Froot\u002Faf_input\u002Ffold_input.json \\\n    --model_dir=\u002Froot\u002Fmodels \\\n    --output_dir=\u002Froot\u002Faf_output\n```\n\n### 3. 关键参数说明\n- `--run_data_pipeline`: (默认 `true`) 是否运行数据流水线（基因序列比对和模板搜索）。此步骤仅消耗 CPU 且耗时较长，若无本地数据库可在此阶段联网搜索，或在无 GPU 机器上预先运行。\n- `--run_inference`: (默认 `true`) 是否运行模型推理。此步骤必须使用 GPU。\n- 若只需测试推理性能且已准备好所有特征数据，可设置 `--run_data_pipeline=false` 以跳过耗时的搜索步骤。\n\n预测完成后，结果文件（包括 PDB 结构和置信度评分）将保存在 `$HOME\u002Faf_output` 目录中。","某生物医药公司的结构生物学团队正在研发一款针对新型病毒蛋白酶的小分子抑制剂，急需解析该蛋白酶与候选药物分子及辅助因子的精确结合构象以指导优化。\n\n### 没有 alphafold3 时\n- **复合物预测能力缺失**：传统工具仅能单独预测蛋白质结构，无法直接模拟蛋白质与小分子配体、核酸或修饰基团的相互作用，必须依赖耗时且昂贵的冷冻电镜实验或复杂的分子对接软件进行拼凑。\n- **精度难以保证**：对于涉及共价修饰或非标准氨基酸的复杂体系，现有计算方法往往误差较大，导致生成的结合口袋模型失真，误导后续的药物化学修饰方向。\n- **研发周期冗长**：从提出假设到获得可信的三维结构模型通常需要数周甚至数月，严重拖慢了先导化合物优化的迭代速度，使团队在激烈的药物竞赛中处于劣势。\n\n### 使用 alphafold3 后\n- **全原子复合物一键生成**：alphafold3 能够直接输入蛋白序列、小分子配体及修饰信息，端到端地输出包含所有生物分子的高精度复合物结构，无需人工拼接不同组件。\n- **相互作用细节清晰可见**：模型精准预测了氢键网络、疏水接触及共价键位置，帮助研究人员直观理解药物分子如何特异性阻断病毒蛋白酶活性，显著提升了结构的可信度。\n- **迭代效率飞跃提升**：原本需要数周的实验验证周期被缩短至数小时，团队可在一天内对数十种衍生物进行虚拟筛选和结构评估，大幅加速了候选药物的定型过程。\n\nalphafold3 通过实现生物分子复合物的高精度端到端预测，将结构驱动的药物研发从“实验试错”时代推向了“计算设计”的新阶段。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_alphafold3_d9339130.jpg","google-deepmind","Google DeepMind","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fgoogle-deepmind_06b1dd17.png","",null,"https:\u002F\u002Fwww.deepmind.com\u002F","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind",[84,88,92,96,100],{"name":85,"color":86,"percentage":87},"Python","#3572A5",85.1,{"name":89,"color":90,"percentage":91},"C++","#f34b7d",14,{"name":93,"color":94,"percentage":95},"Shell","#89e051",0.4,{"name":97,"color":98,"percentage":99},"Dockerfile","#384d54",0.3,{"name":101,"color":102,"percentage":103},"CMake","#DA3434",0.2,7852,1184,"2026-04-18T08:01:47","NOASSERTION",4,"Linux","必需。推理部分 (--run_inference) 需要 GPU。具体型号和显存未说明，但命令示例中包含 '--gpus all'，且依赖 JAX 生态，通常建议使用 NVIDIA GPU。","未说明",{"notes":113,"python":111,"dependencies":114},"1. 官方推荐使用 Docker 容器运行。2. 模型参数不公开下载，需向 Google DeepMind 申请并获得批准后方可使用。3. 运行分为数据流水线（CPU 密集型，可无 GPU）和推理（需 GPU）两个阶段。4. 需自行准备或挂载大型生物数据库（如 BFD, PDB, UniProt 等）。5. 仅限非商业用途（源代码遵循 CC-BY-NC-SA 4.0 协议）。",[115,116,117,118,119,120,121,122,123,124],"Docker","JAX","Haiku (dm-haiku)","NumPy","RDKit","HMMER Suite","DSSP","libcifpp","abseil-cpp\u002Fpy","Tokamax",[18],"2026-03-27T02:49:30.150509","2026-04-19T03:05:11.376277",[129,134,139,144,149,154],{"id":130,"question_zh":131,"answer_zh":132,"source_url":133},41311,"AlphaFold 3 支持哪些 GPU 设备？在 V100 或 RTX 20 系列上运行是否正常？","目前官方仅支持和测试了 A100 和 H100 设备。在 V100、RTX 2060S 和 RTX Quadro 4000 等 CUDA Capability 7.x 的 GPU 上运行时会出现严重问题（输出类似随机噪声），且该问题已确认不会投入更多时间修复。建议仅使用受支持的硬件以获得准确结果。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fissues\u002F59",{"id":135,"question_zh":136,"answer_zh":137,"source_url":138},41312,"如何在 Blackwell 架构（如 RTX Pro 6000）或 CC 12.0 兼容的 GPU 上运行 AlphaFold 3？","Blackwell 兼容性问题已通过更新依赖项并迁移到 Tokamax 解决。请确保更新代码至包含提交 3890782 和 b78e215 之后的版本。此外，需要安装 NVIDIA 驱动版本 575.64.03（CUDA 12.9），Blackwell 必须使用 open driver。若需消除警告，可在 run_alphafold.py 顶部添加许可证声明注释。建议使用 uv 工具管理 Python 环境。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fissues\u002F394",{"id":140,"question_zh":141,"answer_zh":142,"source_url":143},41313,"如何处理共价配体预测中出现的键长不合理或离去原子未移除的问题？","AlphaFold 3 默认会忽略离去原子标记（leaving_atom_flag），导致结构中包含应被移除的原子，建议预处理时手动移除这些原子。对于键长不正确的问题（如配体与氨基酸侧链间的键长过短），目前模型内部难以直接修复。变通方案包括：1. 使用外部软件（如 GROMACS 或 REFMAC5）对生成的结构进行能量最小化；2. 手动构建融合残基 - 配体复合物并生成自定义 CCD CIF 文件输入模型，以获得更合理的初始几何构型。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fissues\u002F159",{"id":145,"question_zh":146,"answer_zh":147,"source_url":148},41314,"能否在天冬氨酸（Aspartic Acid）的酸性基团氧原子上进行共价修饰？","可以直接尝试在 JSON 输入中指定共价连接，但如果出现异常的四面体成键（配体原子同时连接到两个氧原子和中心碳原子），可能是可视化软件（如旧版 PyMOL）的显示问题或输入定义冲突。建议：1. 升级至最新版 PyMOL（如 v3.1.6）查看结构，确认是否仅为显示错误；2. 若需精确控制，可预先修改天冬氨酸残基（例如甲基化），生成自定义的 .cif 文件描述该修饰残基，并在输入的 modifications 字段中引用该自定义组件，而不是直接依赖默认的共价连接逻辑。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fissues\u002F428",{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},41315,"如何预测蛋白质 - 肽复合物结构？","预测蛋白质 - 肽复合物时，不能仅依靠简单的序列输入。需要在输入的 JSON 文件中正确定义肽链作为独立的实体（entity），并明确指定其与蛋白质链之间的相互作用或共价连接（如果存在）。确保在 `sequences` 部分将肽段列为单独条目，并在 `bonded_atom_pairs` 或其他相关字段中正确描述复合物界面或化学键。如果直接修改 JSON 仍失败，请检查是否符合 AlphaFold 3 对多链复合物输入格式的具体要求，特别是链 ID 和残基编号的唯一性。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fissues\u002F135",{"id":155,"question_zh":156,"answer_zh":157,"source_url":138},41316,"在运行 AlphaFold 3 时遇到 JAX 库冲突或依赖项错误怎么办？","针对 Blackwell 等新架构出现的 JAX 冲突，通常是因为依赖包版本过旧。解决方案是更新项目根目录下的 `dev-requirements.txt` 文件，使用更新的包版本（参考社区提供的更新后文件），并重新安装环境。务必确保代码库已更新到最新提交，特别是涉及依赖迁移的提交。使用 `uv` 等现代包管理工具有助于更稳定地解析和安装复杂的依赖关系。",[159,164],{"id":160,"version":161,"summary_zh":162,"released_at":163},333253,"v3.0.1","版本 3.0.1 引入了新功能、提升了性能、改进了文档，并修复了若干 bug。我们感谢所有报告问题、提出新功能建议以及提交 Pull Request 的贡献者！\n\n## 更改日志\n\n以下是部分新功能、性能优化和 bug 修复的摘要。完整的更改日志请参见：https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold3\u002Fcompare\u002Fv3.0.0...v3.0.1。\n\n### 新功能\n\n* 现在可以在不使用模板的情况下运行，但仍会搜索 MSA。\n* 当提供 MSA 时，现在可以执行模板搜索。\n* 现在可以将 MSA 和模板指定为外部文件，而不是内联在输入 JSON 中。\n* 新增以下标志：`--max_template_date`、`--diffusion_num_samples`、`--num_recycles`、`--num_seeds`。\n* 将 `--conformer_max_iterations` 公开为一个标志，以允许增加 RDKit 的迭代次数。\n* 添加了一个选项，用于返回模型嵌入并将其写入文件。\n* 在多 GPU 系统上，现在可以指定使用的 GPU。\n* 输出的 mmCIF 文件中现在包含了每个残基的 pLDDT 值。\n\n### 性能改进\n\n* 用一个显著更快的数据库下载脚本替换了原有的脚本，并添加了 GCP 后处理脚本。\n* 改进了 Stockholm 转换器，在转换为 A3M 时不会读取不必要的序列。\n* 进一步优化了从 Stockholm 到 A3M 的转换流程，避免将整个文件加载到内存中。\n* 通过不再读取和解析无关的 mmCIF 文件，加快了模板搜索的速度。如果 PDB 存储在速度较慢的文件系统上，这一改进尤为显著。\n* 对输入 JSON 的解析进行了重构，改为使用迭代器方式，从而避免一次性将所有输入加载到内存中导致 OOM。\n* 在输出 JSON 时增加了序列去重功能，使得同源寡聚体的输出文件更小。\n\n### Bug 修复\n\n* 如果提供了空的 MSA 和模板，数据管道现在会被跳过。\n* 静默了无关的 Numpy 警告信息。\n* 改进了 Singularity 的安装说明。\n* 修复了 Pallas GLU 内核在处理大型输入时的问题，并更新了编译桶的相关文档。\n* 修正了设置，使其能够兼容 Python 3.12。\n* 如果输出目录已存在，AlphaFold 不会覆盖其中的文件，而是会创建一个新的目录。\n* 将 DATIVE 键类型转换为 SINGLE，以匹配训练过程中使用的键类型。\n* 在嵌入名称中将原子元素名统一为大写，以与训练时保持一致。\n* 修复了解压 PDB mmCIF 文件时的用户权限问题。\n* 在获取生成的构象时，明确使用 `conformer_id`。\n* 修复了当描述中包含制表符时 A3M 到 Stockholm 的转换问题。\n* 对 glob 返回的输入文件进行排序——否则文件顺序是任意的。\n* 确保用户 CCD 包含所有必要字段，并改进了文档。\n* 修复了 SMILES 配体中对双字母原子的错误处理。\n* 在可用的情况下，使用 OpenEye 的规范 SMILES 表示。\n* 如果无法创建分子定义，则抛出清晰的错误信息。\n* 为 CUDA Capability 7.x 的 GPU 添加了 XLA 标志的 workaround。\n* 对 Pallas 内核中处理 PID 时的数值精度问题进行了小幅修复。\n* 正确处理单字母序列中的非标准残基。","2025-01-23T09:15:06",{"id":165,"version":166,"summary_zh":167,"released_at":168},333254,"v3.0.0","AlphaFold 3 初次发布。\n\nPiperOrigin-RevId: 695257954\n","2024-11-11T10:49:52"]