[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tool-ai-boost--awesome-ai-for-science":3,"similar-ai-boost--awesome-ai-for-science":47},{"id":4,"github_repo":5,"name":6,"description_en":7,"description_zh":8,"ai_summary_zh":8,"readme_en":9,"readme_zh":10,"quickstart_zh":11,"use_case_zh":12,"hero_image_url":13,"owner_login":14,"owner_name":15,"owner_avatar_url":16,"owner_bio":17,"owner_company":17,"owner_location":17,"owner_email":17,"owner_twitter":18,"owner_website":19,"owner_url":20,"languages":17,"stars":21,"forks":22,"last_commit_at":23,"license":24,"difficulty_score":25,"env_os":26,"env_gpu":27,"env_ram":27,"env_deps":28,"category_tags":31,"github_topics":33,"view_count":41,"oss_zip_url":17,"oss_zip_packed_at":17,"status":42,"created_at":43,"updated_at":44,"faqs":45,"releases":46},2088,"ai-boost\u002Fawesome-ai-for-science","awesome-ai-for-science","A curated list of awesome AI tools, libraries, papers, datasets, and frameworks that accelerate scientific discovery — from physics and chemistry to biology, materials, and beyond.","awesome-ai-for-science 是一个专为加速科学发现而打造的精选资源库，汇集了全球范围内优质的 AI 工具、代码库、学术论文、数据集及框架。它致力于解决科研人员在跨学科研究中面临的“资源分散”与“技术门槛”难题，将原本零散分布在物理、化学、生物、材料科学乃至天文学等领域的 AI 应用成果进行了系统化梳理。\n\n无论是希望利用人工智能优化实验流程的科研人员，还是想要开发垂直领域应用的开发者，都能在这里找到得力助手。该资源库不仅涵盖了从文献管理、数据可视化到自动化科研工作流的全套工具链，还特别突出了\"AI for Science\"的前沿趋势，如科学基础模型、论文转代码复现以及自主研究智能体等独特技术方向。通过提供按学科分类的详细指南和经过验证的开源项目，awesome-ai-for-science 有效降低了科学家使用 AI 技术的门槛，促进了计算机科学与传统自然科学的深度融合，是推动现代科研范式变革的重要基础设施。","\u003Cdiv align=\"center\">\n  \u003Ch1>✨ Awesome AI for Science (AI4Science) ✨\u003C\u002Fh1>\n  \n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fai-boost_awesome-ai-for-science_readme_9bb893d60e49.jpg\" alt=\"Awesome AI for Science Banner\" width=\"100%\">\n  \n  \u003Cp align=\"center\">\n    A curated list of awesome AI tools, libraries, papers, datasets, and frameworks that accelerate \u003Cstrong>scientific discovery\u003C\u002Fstrong> across all disciplines.\n  \u003C\u002Fp>\n  \n  \u003C!-- Keep these links. Translations will automatically update with the README. -->\n  \u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fde\u002Fai-boost\u002Fawesome-ai-for-science\">Deutsch\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fen\u002Fai-boost\u002Fawesome-ai-for-science\">English\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fes\u002Fai-boost\u002Fawesome-ai-for-science\">Español\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Ffr\u002Fai-boost\u002Fawesome-ai-for-science\">français\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fja\u002Fai-boost\u002Fawesome-ai-for-science\">日本語\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fko\u002Fai-boost\u002Fawesome-ai-for-science\">한국어\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fpt\u002Fai-boost\u002Fawesome-ai-for-science\">Português\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fru\u002Fai-boost\u002Fawesome-ai-for-science\">Русский\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fzh\u002Fai-boost\u002Fawesome-ai-for-science\">中文\u003C\u002Fa>\n  \u003C\u002Fp>\n  \n  \u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fawesome.re\">\u003Cimg src=\"https:\u002F\u002Fawesome.re\u002Fbadge.svg\" alt=\"Awesome\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicense\u002FMIT\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow.svg\" alt=\"License: MIT\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fai-boost\u002Fawesome-ai-for-science\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fai-boost\u002Fawesome-ai-for-science.svg?style=social&label=Star\" alt=\"GitHub stars\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fai-boost\u002Fawesome-ai-for-science\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fforks\u002Fai-boost\u002Fawesome-ai-for-science.svg?style=social&label=Fork\" alt=\"GitHub forks\">\u003C\u002Fa>\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n> AI is revolutionizing scientific research - from drug discovery and materials design to climate modeling and astrophysics. This repository collects the best resources to help researchers leverage AI in their work.\n\n## 📚 Contents\n\n- [🧪 AI Tools for Research](#-ai-tools-for-research)\n- [📄 Paper→Poster \u002F Slides \u002F Graphical Abstract](#-paperposter--slides--graphical-abstract)\n- [📊 Chart Understanding & Generation](#-chart-understanding--generation)\n- [🔄 Paper-to-Code & Reproducibility](#-paper-to-code--reproducibility)\n- [📋 Scientific Documentation & Parsing](#-scientific-documentation--parsing)\n- [🧰 Research Workbench & Plugins](#-research-workbench--plugins)\n- [🕸 Knowledge Extraction & Scholarly KGs](#-knowledge-extraction--scholarly-kgs)\n- [🤖 Research Agents & Autonomous Workflows](#-research-agents--autonomous-workflows)\n- [🏷 Data Labeling & Curation](#-data-labeling--curation)\n- [⚗ Scientific Machine Learning](#-scientific-machine-learning)\n- [📖 Papers & Reviews](#-papers--reviews)\n- [🔬 Domain-Specific Applications](#-domain-specific-applications)\n  - [🧬 Biology & Medicine](#-biology--medicine)\n  - [⚛ Chemistry & Materials](#-chemistry--materials)  \n  - [🌌 Physics & Astronomy](#-physics--astronomy)\n  - [🌍 Earth & Climate Science](#-earth--climate-science)\n  - [🌾 Agriculture & Ecology](#-agriculture--ecology)\n- [🤖 Foundation Models for Science](#-foundation-models-for-science)\n- [📈 Datasets & Benchmarks](#-datasets--benchmarks)\n- [💻 Computing Frameworks](#-computing-frameworks)\n- [🎓 Educational Resources](#-educational-resources)\n- [🏛 Research Communities](#-research-communities)\n- [📚 Related Awesome Lists](#-related-awesome-lists)\n\n---\n\n## 🧪 AI Tools for Research\n\n### Literature & Knowledge Management\n- [Semantic Scholar](https:\u002F\u002Fwww.semanticscholar.org\u002F) - AI-powered academic search (Allen AI)\n- [arXiv](https:\u002F\u002Farxiv.org\u002F) - Open-access repository of electronic preprints and postprints\n- [OpenAlex](https:\u002F\u002Fopenalex.org\u002F) - Open catalog of scholarly papers and authors\n- [CORE](https:\u002F\u002Fcore.ac.uk\u002F) - Aggregator of open access research papers\n\n### Data Analysis & Visualization\n- [PandasAI](https:\u002F\u002Fgithub.com\u002FSinaptik-AI\u002Fpandas-ai) - Conversational data analysis using natural language\n- [DeepAnalyze](https:\u002F\u002Fgithub.com\u002Fruc-datalab\u002FDeepAnalyze) - First agentic LLM for autonomous data science with end-to-end pipeline from data to analyst-grade reports\n- [AutoViz](https:\u002F\u002Fgithub.com\u002FAutoViML\u002FAutoViz) - Automated data visualization with minimal code\n- [Chat2Plot](https:\u002F\u002Fgithub.com\u002Fnyanp\u002Fchat2plot) - Secure text-to-visualization through standardized chart specifications\n\n### Data Labeling & Annotation\n- [Label Studio](https:\u002F\u002Fgithub.com\u002Fheartexlabs\u002Flabel-studio) - Multi-type data labeling and annotation tool\n- [Snorkel](https:\u002F\u002Fgithub.com\u002Fsnorkel-team\u002Fsnorkel) - Programmatic data labeling and weak supervision\n\n### Research Workbench & Plugins\n- [Claude Scientific Skills](https:\u002F\u002Fgithub.com\u002FK-Dense-AI\u002Fclaude-scientific-skills) - Comprehensive collection of 125+ ready-to-use scientific skill modules for Claude AI across bioinformatics, cheminformatics, clinical research, ML, and materials science\n\n---\n\n## 📄 Paper→Poster \u002F Slides \u002F Graphical Abstract\n\n### Poster Generation\n- [Paper2Poster](https:\u002F\u002Fgithub.com\u002FPaper2Poster\u002FPaper2Poster) - Multi-agent system with Parser-Planner-Painter architecture converting `paper.pdf` to editable `poster.pptx`, outperforms GPT-4o with 87% fewer tokens\n- [mPLUG-PaperOwl](https:\u002F\u002Fgithub.com\u002FX-PLUG\u002FmPLUG-DocOwl) - Multimodal LLM for scientific charts and diagrams understanding\u002Fgeneration\n\n### Slides & Presentation Generation\n- [Auto-Slides](https:\u002F\u002Fauto-slides.github.io\u002F) - Multi-agent academic paper to high-quality presentation slides with interactive refinement\n- [PPTAgent](https:\u002F\u002Fgithub.com\u002Ficip-cas\u002FPPTAgent) - Beyond text-to-slides generation with PPTEval multi-dimensional evaluation (EMNLP 2025)\n- [paper2slides](https:\u002F\u002Fgithub.com\u002Ftakashiishida\u002Fpaper2slides) - Transform arXiv papers into Beamer slides using LLMs\n- [PaperToSlides](https:\u002F\u002Fgithub.com\u002Fjxtse\u002FPaperToSlides) - AI-powered tool that automatically converts academic papers (PDF) into presentation slides\n- [pdf2slides](https:\u002F\u002Fgithub.com\u002Fha0ranyu\u002Fpdf2slides) - Convert PDF files into editable slides with three lines of code\n- [SlideDeck AI](https:\u002F\u002Fgithub.com\u002Fbarun-saha\u002Fslide-deck-ai) - Co-create PowerPoint presentations with Generative AI from documents or topics\n- [AI Multi-Agent Presentation Builder](https:\u002F\u002Fgithub.com\u002FAzure-Samples\u002Fai-multi-agent-presentation-builder) - Azure Semantic Kernel multi-agent PPT generation reference\n\n### Video & Media Generation\n- [Paper2Video](https:\u002F\u002Fgithub.com\u002Fshowlab\u002FPaper2Video) - First benchmark for automatic video generation from scientific papers (NeurIPS 2025)\n- [paper2video](https:\u002F\u002Fgithub.com\u002Fmett29\u002Fpaper2video) - Transform arXiv research papers into engaging presentations and YouTube-ready videos\n\n### Website & Interactive Content Generation\n- [Paper2All](https:\u002F\u002Fgithub.com\u002FYuhangChen1\u002FPaper2All) - AI-powered pipeline converting papers into interactive websites, posters, and multimedia presentations with \"Let's Make Your Paper Alive!\" philosophy\n\n### Chart & Visualization Generation  \n*Note: For comprehensive chart understanding and code generation tools, see [📊 Chart Understanding & Generation](#-chart-understanding--generation) section*\n\n---\n\n## 📊 Chart Understanding & Generation\n\n### Chart-to-Code & Reproducibility\n- [ChartCoder (ACL 2025)](https:\u002F\u002Faclanthology.org\u002F2025.acl-long.363\u002F) - Multimodal LLM for chart-to-code generation, 7B model outperforms larger open-source MLLMs\n- [ChartAssistant \u002F ChartAst (ACL 2024)](https:\u002F\u002Fgithub.com\u002FOpenGVLab\u002FChartAst) - Universal chart comprehension and reasoning model\n- [Chart-to-Text Datasets](https:\u002F\u002Fgithub.com\u002Fvis-nlp\u002FChart-to-text) - Large-scale chart summarization datasets for training chart description capabilities\n\n### Scientific Visualization Tools\n- [Chat2Plot](https:\u002F\u002Fgithub.com\u002Fnyanp\u002Fchat2plot) - Secure text-to-visualization through standardized chart specifications\n- [AutoViz](https:\u002F\u002Fgithub.com\u002FAutoViML\u002FAutoViz) - Automated data visualization with minimal code\n- [PlotlyAI](https:\u002F\u002Fplotly.com\u002Fai\u002F) - AI-powered data visualization and dashboard creation\n\n---\n\n## 🔄 Paper-to-Code & Reproducibility\n\n### Automated Code Generation\n- [AutoP2C](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.20115) - LLM agent framework generating runnable repositories from academic papers\n- [ResearchCodeAgent](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.20117) - Multi-agent system for automated codification of research methodologies\n- [ToolMaker](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2502.11705) - Convert papers with code into callable agent tools\n\n### Experiment Automation\n- [BioProBench](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FGreatCaptainNemo\u002FBioProBench) - Comprehensive benchmark for automatic evaluation of LLMs on biological protocols and procedural understanding\n- [Alhazen](https:\u002F\u002Fchanzuckerberg.github.io\u002Falhazen\u002F) - Extract experimental metadata and protocol information from scientific documents\n\n---\n\n## 📋 Scientific Documentation & Parsing\n\n### High-Performance Document Processing\n- [MinerU (2024\u002F2025)](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FMinerU) - SOTA multimodal document parsing with 1.2B parameters outperforming GPT-4o, converts PDFs to LLM-ready Markdown\u002FJSON\n- [PDF-Extract-Kit (2024)](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FPDF-Extract-Kit) - Comprehensive toolkit for high-quality PDF content extraction with layout detection, formula recognition, and OCR\n- [Docling (IBM, AAAI 2025)](https:\u002F\u002Fresearch.ibm.com\u002Fpublications\u002Fdocling-an-efficient-open-source-toolkit-for-ai-driven-document-conversion) - Multi-format (PDF\u002FDOCX\u002FPPTX\u002FHTML\u002FImages) → structured data (Markdown\u002FJSON) with layout reconstruction, table\u002Fformula recovery\n- [Nougat (Meta AI)](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fnougat) - Neural optical understanding for academic documents, transforms scientific PDFs to Markdown with mathematical formula support\n- [PaddleOCR 3.0 (2024\u002F2025)](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR) - Advanced OCR with PP-StructureV3 document parsing, 13% accuracy improvement, supports 80+ languages\n- [Unstructured](https:\u002F\u002Fgithub.com\u002FUnstructured-IO\u002Funstructured) - Production-grade ETL for transforming complex documents into structured formats, with open-source API\n- [Marker](https:\u002F\u002Fgithub.com\u002Fdatalab-to\u002Fmarker) - High-accuracy PDF→Markdown\u002FJSON\u002FHTML conversion, specialized for tables\u002Fformulas\u002Fcode blocks with benchmark scripts\n- [S2ORC doc2json (AllenAI)](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fs2orc-doc2json) - Large-scale PDF\u002FLaTeX\u002FJATS parsing to standardized JSON for millions of papers\n- [GROBID](https:\u002F\u002Fgithub.com\u002Fkermitt2\u002Fgrobid) - Machine learning software for extracting structured metadata from scholarly documents\n- [Science-Parse \u002F SPv2 (AllenAI)](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fscience-parse) - Parse scientific papers to structured fields (title\u002Fauthor\u002Fsections\u002Freferences)\n\n### Production Pipelines & Data Preparation\n- [IBM Data Prep Kit: PDF→Parquet](https:\u002F\u002Fibm.github.io\u002Fdata-prep-kit\u002Ftransforms\u002Flanguage\u002Fpdf2parquet\u002F) - Large-scale scientific document ingestion pipeline with optimization configurations\n- [Mozilla document-to-markdown](https:\u002F\u002Fgithub.com\u002Fmozilla-ai\u002Fdocument-to-markdown) - Docling-powered parsing with UI\u002FCLI demonstration for rapid prototyping\n\n### Figure & Table Extraction\n- [PDFFigures2](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fpdffigures2) - Extract figures, tables, captions, and section titles from scholarly PDFs\n- [TableBank](https:\u002F\u002Fgithub.com\u002Fdoc-analysis\u002FTableBank) - Large-scale table detection and recognition dataset with pre-trained models\n\n### Scientific Literature RAG & Analysis\n- [PaperQA2](https:\u002F\u002Fgithub.com\u002Ffuture-house\u002Fpaper-qa) - High-accuracy RAG for scientific PDFs with citation support, agentic RAG, and contradiction detection\n- [OpenScholar](https:\u002F\u002Fgithub.com\u002FAkariAsai\u002FOpenScholar) - Retrieval-augmented LM synthesizing scientific literature from 45M papers with human-expert-level citation accuracy, outperforming GPT-4o by 5% on ScholarQABench (Nature 2026, UW & Ai2)\n- [paper-reviewer](https:\u002F\u002Fgithub.com\u002Fdeep-diver\u002Fpaper-reviewer) - Generate comprehensive reviews from arXiv papers and convert to blog posts\n\n---\n\n## 🧰 Research Workbench & Plugins\n\n### Interactive Research Environments\n- [Jupyter AI (JupyterLab Extension)](https:\u002F\u002Fgithub.com\u002Fjupyterlab\u002Fjupyter-ai) - Official Jupyter extension with `%%ai` magic commands and sidebar chat assistant, connecting multiple model providers and local inference\n- [Notebook Intelligence (NBI)](https:\u002F\u002Fgithub.com\u002Fnotebook-intelligence\u002Fnotebook-intelligence) - AI coding assistant for JupyterLab with agent mode, supporting arbitrary LLM providers (2025+)\n- [Google Colab AI Features](https:\u002F\u002Fcolab.research.google.com\u002F) - Integrated AI assistance for data science and research notebooks\n\n### Literature Management Plugins\n- [PapersGPT for Zotero](https:\u002F\u002Fgithub.com\u002Fpapersgpt\u002Fpapersgpt-for-zotero) - Multi-PDF conversation, retrieval, and citation in Zotero with commercial\u002Flocal models (Ollama), MCP support\n- [Zotero-GPT (MuiseDestiny)](https:\u002F\u002Fgithub.com\u002FMuiseDestiny\u002Fzotero-gpt) - Classic open-source plugin for document Q&A and summarization within Zotero\n- [Better BibTeX for Zotero](https:\u002F\u002Fretorque.re\u002Fzotero-better-bibtex\u002F) - Enhanced citation key management and LaTeX integration\n\n### Scientific Writing & Collaboration\n- [Notion AI](https:\u002F\u002Fwww.notion.so\u002Fproduct\u002Fai) - AI-powered research note-taking and knowledge management\n- [Obsidian Smart Connections](https:\u002F\u002Fgithub.com\u002Fbrianpetro\u002Fobsidian-smart-connections) - AI-powered note linking and research graph navigation\n- [Research Rabbit](https:\u002F\u002Fwww.researchrabbit.ai\u002F) - AI-powered literature discovery and research network mapping\n\n---\n\n## 🕸 Knowledge Extraction & Scholarly KGs\n\n### Knowledge Graph Construction\n- [iText2KG](https:\u002F\u002Fgithub.com\u002FAuvaLab\u002Fitext2kg) - Incremental knowledge graph construction using LLMs with entity extraction and Neo4j visualization\n- [GraphGen](https:\u002F\u002Fgithub.com\u002Fopen-sciencelab\u002FGraphGen) - Knowledge graph-guided synthetic data generation for LLM fine-tuning, achieving strong performance on scientific QA (GPQA-Diamond) and math reasoning (AIME)\n- [KoPA](https:\u002F\u002Fgithub.com\u002Fzjukg\u002FKoPA) - Structure-aware prefix adaptation for integrating LLMs with knowledge graphs (ACM MM 2024)\n- [Scholarly KGQA](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.09841) - LLM-powered question answering over scholarly knowledge graphs (ArXiv paper)\n\n### Knowledge Graph Resources\n- [Awesome-LLM-KG](https:\u002F\u002Fgithub.com\u002FRManLuo\u002FAwesome-LLM-KG) - Comprehensive collection of papers on unifying LLMs and knowledge graphs\n\n---\n\n## 🤖 Research Agents & Autonomous Workflows\n\n### Autonomous Research Systems (2024-2025 Breakthroughs)\n- [The AI Scientist v1 (2024)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.06292) - First fully autonomous research system: hypothesis→experiment→writing→review simulation\n- [The AI Scientist v2 (2025)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.08066) - Enhanced with Agentic Tree Search, reduced template dependency, first workshop-level accepted paper\n- [DeepScientist](https:\u002F\u002Fgithub.com\u002FResearAI\u002FDeepScientist) - First system progressively surpassing human SOTA on frontier AI tasks (183.7%, 1.9%, 7.9% improvements), month-long autonomous discovery with 20,000+ GPU hours\n- [Kosmos](https:\u002F\u002Fgithub.com\u002Fjimmc414\u002FKosmos) - Extended autonomy AI scientist with 200 parallel agent rollouts, 42K lines of code execution, 1.5K papers analyzed per run, achieving 79.4% accuracy and 7 scientific discoveries (Edison Scientific)\n- [AlphaResearch](https:\u002F\u002Fgithub.com\u002Fanswers111\u002Falpha-research) - Autonomous algorithm discovery combining evolutionary search with peer-review reward models, achieving best-known performance on circle packing problems\n- [AI-Researcher](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FAI-Researcher) - Autonomous pipeline from literature review→hypothesis→algorithm implementation→publication-level writing with Scientist-Bench evaluation\n- [Agent Laboratory](https:\u002F\u002Fagentlaboratory.github.io\u002F) - Multi-agent workflows for complete research cycles with AgentRxiv for cumulative discovery\n- [InternAgent](https:\u002F\u002Fgithub.com\u002FAlpha-Innovator\u002FInternAgent) - Closed-loop multi-agent system from hypothesis to verification across 12 scientific tasks, #1 on MLE-Bench (36.44%)\n- [freephdlabor](https:\u002F\u002Fgithub.com\u002Fltjed\u002Ffreephdlabor) - First fully customizable open-source multiagent framework automating complete research lifecycle from idea conception to LaTeX papers with dynamic workflows\n- [ToolUniverse](https:\u002F\u002Fgithub.com\u002Fmims-harvard\u002FToolUniverse) - Democratizing AI scientists by transforming any LLM into research systems with 600+ scientific tools (Harvard MIMS)\n- [LabClaw](https:\u002F\u002Fgithub.com\u002Fwu-yc\u002FLabClaw) - Skill operating layer for biomedical AI agents with 211 production-ready SKILL.md files across 7 domains (biology, pharmacology, medicine, data science, literature search), enabling modular dry-lab reasoning and protocol composition for Stanford LabOS-compatible agents\n- [Robin](https:\u002F\u002Fgithub.com\u002FFuture-House\u002Frobin) - FutureHouse's end-to-end scientific discovery multi-agent system orchestrating literature search (Crow\u002FFalcon) and data analysis (Finch) agents, first AI-generated drug discovery identifying ripasudil as novel dry AMD therapeutic (2025)\n- [Aviary](https:\u002F\u002Fgithub.com\u002FFuture-House\u002Faviary) - Language agent gymnasium for challenging scientific tasks including DNA manipulation, literature search, and protein engineering\n- [Curie](https:\u002F\u002Fgithub.com\u002FJust-Curieous\u002FCurie) - Automated and rigorous experiments using AI agents for scientific discovery\n- [POPPER](https:\u002F\u002Fgithub.com\u002Fsnap-stanford\u002FPOPPER) - Automated hypothesis testing with agentic sequential falsifications\n- [autoresearch](https:\u002F\u002Fgithub.com\u002Fkarpathy\u002Fautoresearch) - Andrej Karpathy's autonomous LLM research framework: AI agent runs overnight experiments on a real training setup, auto-editing code→5min training→evaluation in a loop, ~100 experiments per night on a single GPU\n- [UniScientist](https:\u002F\u002Fgithub.com\u002FUniPat-AI\u002FUniScientist) - Universal scientific research intelligence covering 50+ disciplines, repositioning LLMs as cross-disciplinary generators with human experts as verifiers; 30B model outperforms Claude Opus and GPT on 5 research benchmarks\n\n### Evaluation & Benchmarking\n- [ScienceAgentBench (ICLR 2025)](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002FScienceAgentBench) - 102 executable tasks from 44 peer-reviewed papers across 4 disciplines with containerized evaluation\n- [BuildArena](https:\u002F\u002Fgithub.com\u002FAI4Science-WestlakeU\u002FBuildArena) - First physics-aligned interactive benchmark for LLM agents in engineering construction, designing rockets\u002Fcars\u002Fbridges in physics simulator with 3D spatial geometry library\n- [SciTrust (2024)](https:\u002F\u002Fimpact.ornl.gov\u002Fen\u002Fpublications\u002Fscitrust-evaluating-the-trustworthiness-of-large-language-models-) - Trustworthiness evaluation framework for scientific LLMs (truthfulness, hallucination, sycophancy)\n- [SciCode](https:\u002F\u002Fgithub.com\u002Fscicode-bench\u002FSciCode) - Research coding benchmark curated by scientists with 338 subproblems across 16 subdomains (physics, math, materials, biology, chemistry), evaluating LLMs on realistic scientific programming tasks with gold-standard solutions (NeurIPS 2024)\n- [SciBench](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.10635) - College-level scientific problem-solving evaluation across multiple domains\n\n### Academic Review & Evaluation\n- [AgentReview](https:\u002F\u002Fagentreview.github.io\u002F) - LLM agents simulating academic peer review ecosystems\n- [LLM-Peer-Review](https:\u002F\u002Fgithub.com\u002FVijayGKR\u002FLLM-Peer-Review) - Web application for LLM-assisted manuscript review and annotation\n\n### Domain-Specific Research Agents\n- [Aletheia](https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.10177) - Google DeepMind's autonomous mathematics research agent powered by Gemini Deep Think, autonomously solving 4 open problems from 700 Erdős conjectures and generating complete research papers without human intervention (February 2026)\n- [AlphaGeometry](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphageometry) - DeepMind's Olympiad-level geometry theorem prover combining neural language model with symbolic deduction engine, AlphaGeometry2 solves 84% of IMO geometry problems (42\u002F50) at gold-medalist level (Nature 2024)\n- [Goedel-Prover-V2](https:\u002F\u002Fgithub.com\u002FGoedel-LM\u002FGoedel-Prover-V2) - Strongest open-source automated theorem prover in Lean 4, 8B model matches DeepSeek-Prover-V2-671B at 84.6% MiniF2F, 32B model achieves 90.4% with self-correction, using scaffolded data synthesis and verifier-guided proof refinement (Princeton, 2025)\n- [BioDiscoveryAgent](https:\u002F\u002Fgithub.com\u002Fsnap-stanford\u002FBioDiscoveryAgent) - AI agent for biological discovery and research automation\n- [MOOSE](https:\u002F\u002Fgithub.com\u002FZonglinY\u002FMOOSE) - Large Language Models for automated open-domain scientific hypotheses discovery (ACL 2024, ICML Best Poster)\n- [ChemCrow](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.05376) - LLM agents for chemistry research with tool integration\n- [Coscientist](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-023-06792-1) - Autonomous chemical experiment planning and execution\n\n---\n\n## 🏷 Data Labeling & Curation\n\n### Weak Supervision & Auto-Labeling\n- [Snorkel](https:\u002F\u002Fgithub.com\u002Fsnorkel-team\u002Fsnorkel) - Programmatic data labeling and weak supervision for scientific datasets\n- [PandasAI](https:\u002F\u002Fgithub.com\u002FSinaptik-AI\u002Fpandas-ai) - Conversational data analysis and visualization using natural language\n\n---\n\n## ⚗ Scientific Machine Learning\n\n### Neural Differential Equations\n- [torchdiffeq](https:\u002F\u002Fgithub.com\u002Frtqichen\u002Ftorchdiffeq) - PyTorch implementation of neural ODEs\n- [torchdyn](https:\u002F\u002Fgithub.com\u002FDiffEqML\u002Ftorchdyn) - Neural differential equations in PyTorch\n- [diffrax](https:\u002F\u002Fgithub.com\u002Fpatrick-kidger\u002Fdiffrax) - Numerical differential equation solving in JAX\n- [DifferentialEquations.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FDifferentialEquations.jl) - Julia differential equations suite\n- [DiffEqFlux.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FDiffEqFlux.jl) - Neural differential equations in Julia\n\n### Physics-Informed Neural Networks\n- [DeepXDE](https:\u002F\u002Fgithub.com\u002Flululxvi\u002Fdeepxde) - Deep learning library for solving PDEs\n- [Lang-PINN](https:\u002F\u002Fopenreview.net\u002Fforum?id=ONEyVpgK34) - LLM-driven multi-agent system that builds trainable PINNs from natural language task descriptions, achieving 3-5 orders of magnitude MSE reduction and 50%+ execution success improvement (ICLR 2026)\n- [PINNs](https:\u002F\u002Fgithub.com\u002Fmaziarraissi\u002FPINNs) - Physics-informed neural networks\n- [NVIDIA PhysicsNeMo](https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fphysicsnemo) - Open-source framework for building physics-ML models at scale (renamed from Modulus, 2025)\n- [PINA](https:\u002F\u002Fgithub.com\u002FmathLab\u002FPINA) - Physics-Informed Neural networks for Advanced modeling in PyTorch\n- [SciANN](https:\u002F\u002Fgithub.com\u002Fsciann\u002Fsciann) - Keras-based scientific neural networks\n- [NeuralPDE.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FNeuralPDE.jl) - Physics-informed neural networks in Julia\n\n### Neural Operators & Model Discovery\n- [DeepONet](https:\u002F\u002Fgithub.com\u002Flululxvi\u002Fdeeponet) - Learning nonlinear operators\n- [PySINDy](https:\u002F\u002Fgithub.com\u002Fdynamicslab\u002Fpysindy) - Sparse identification of nonlinear dynamics\n- [PySR](https:\u002F\u002Fgithub.com\u002FMilesCranmer\u002FPySR) - High-performance symbolic regression for discovering interpretable scientific equations from data, multi-population evolutionary search with Python\u002FJulia backend, widely used in physics and astronomy (Cambridge, NeurIPS 2023)\n- [LLM-SR](https:\u002F\u002Fgithub.com\u002Fdeep-symbolic-mathematics\u002FLLM-SR) - Scientific equation discovery and symbolic regression using LLMs, combining code generation with evolutionary search (ICLR 2025 Oral)\n- [Fourier Neural Operator](https:\u002F\u002Fgithub.com\u002Fneuraloperator\u002Fneuraloperator) - Learning operators in Fourier space\n\n---\n\n## 📖 Papers & Reviews\n\n### Foundational Papers\n- [Machine Learning for Scientometric Analysis](https:\u002F\u002Farxiv.org\u002Fabs\u002F2109.10073) (2021.09) - Comprehensive review\n- [AI for Science: Progress and Challenges](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.04346) (2023.03) - State of the field\n- [Foundation Models for Science](https:\u002F\u002Farxiv.org\u002Fabs\u002F2205.15075) (2022.05) - Large models in research\n- [Neural Ordinary Differential Equations](https:\u002F\u002Farxiv.org\u002Fabs\u002F1806.07366) (2018.06) - Breakthrough in neural ODEs\n- [Physics-Informed Neural Networks](https:\u002F\u002Farxiv.org\u002Fabs\u002F1711.10561) (2017.11) - Physics-constrained deep learning\n- [Scientific Discovery in the Age of Artificial Intelligence](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-023-06221-2) - Nature review on AI's role in science\n\n### 📊 Comprehensive Surveys & Reviews (2024-2025)\n\n#### AI for Scientific Research\n- [A Survey on AI-assisted Scientific Discovery](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.05151) (2025.02) - Comprehensive overview of LLMs in scientific research lifecycle from literature search to peer review\n- [AI4Research: A Survey of Artificial Intelligence for Scientific Research](https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.01903) (2025.07) - Systematic taxonomy of AI in research\n- [Artificial Intelligence for Science in Quantum, Atomistic, and Continuum Systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.08423) (2023.07) - Unified technical survey across scientific scales with 63 contributors\n- [From Automation to Autonomy: A Survey on Large Language Models in Scientific Discovery](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.13259) (2025.05) - Three-level taxonomy (Tool, Analyst, Scientist)\n- [From AI for Science to Agentic Science: A Survey on Autonomous Scientific Discovery](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.14111) (2025.08) - Comprehensive survey on agentic science across life sciences, chemistry, materials, and physics\n- [Agentic AI for Scientific Discovery: A Survey of Progress, Challenges, and Future Directions](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.08979) (2025.03) - Comprehensive review of AI agents in science\n- [Towards Scientific Intelligence: A Survey of LLM-based Scientific Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.24047) (2025.03) - Scientific AI agent systems\n\n#### Scientific Large Language Models  \n- [A Comprehensive Survey of Scientific Large Language Models and Their Applications](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.10833) (2024.06) - 260+ scientific LLMs across domains\n- [A Survey of Scientific Large Language Models: From Data Foundations to Agent Frontiers](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.21148) (2025.08) - Data-centric view of scientific LLMs\n- [Scientific Large Language Models: A Survey on Biological & Chemical Domains](https:\u002F\u002Farxiv.org\u002Fabs\u002F2401.14656) (2024.01) - Domain-specific scientific LLMs\n\n#### Scientific Machine Learning\n- [Scientific Machine Learning through Physics-Informed Neural Networks: Where we are and What's next](https:\u002F\u002Farxiv.org\u002Fabs\u002F2201.05624) (2022.01) - Comprehensive PINN review\n- [Physics-Informed Neural Networks and Extensions](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.16806) (2024.08) - Recent PINN advances and variants\n- [The frontier of simulation-based inference](https:\u002F\u002Fwww.pnas.org\u002Fdoi\u002F10.1073\u002Fpnas.1912789117) (PNAS 2020) - Foundational review on SBI for scientific computing by Cranmer et al.\n- [From Theory to Application: A Practical Introduction to Neural Operators in Scientific Computing](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.05598) (2025.03) - Implementation-focused guide to DeepONet, FNO, and PCANet\n- [Architectures, variants, and performance of neural operators: A comparative review](https:\u002F\u002Fwww.sciencedirect.com\u002Fscience\u002Farticle\u002Fabs\u002Fpii\u002FS0925231225011907) (2025) - Systematic analysis of DeepONets, integral kernel operators, and transformer-based neural operators\n- [Foundation Models for Environmental Science: A Survey](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.04280) (2025.04) - Environmental applications\n- [Foundation Models in Bioinformatics](https:\u002F\u002Facademic.oup.com\u002Fnsr\u002Farticle\u002F12\u002F4\u002Fnwaf028\u002F7979309) - Biological foundation models\n- [Foundation Models for Materials Discovery](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41524-025-01538-0) (2025) - Perspective on materials AI\n\n#### Uncertainty Quantification\n- [Uncertainty quantification in scientific machine learning: Methods, metrics, and comparisons](https:\u002F\u002Fwww.sciencedirect.com\u002Fscience\u002Farticle\u002Fabs\u002Fpii\u002FS0021999122009652) (J. Comput. Phys. 2023) - Comprehensive framework for UQ in PINNs and neural operators by Psaros et al.\n- [A Survey on Uncertainty Quantification Methods for Deep Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2302.13425) (2023) - Systematic taxonomy of UQ methods from uncertainty source perspective\n\n#### Automation & Self-Driving Laboratories\n- [Self-Driving Laboratories for Chemistry and Materials Science](https:\u002F\u002Fpubs.acs.org\u002Fdoi\u002F10.1021\u002Facs.chemrev.4c00055) (Chem. Rev. 2024) - Comprehensive 100-page review on SDL technology, applications, and infrastructure\n- [Autonomous 'self-driving' laboratories: a review of technology and policy implications](https:\u002F\u002Froyalsocietypublishing.org\u002Fdoi\u002F10.1098\u002Frsos.250646) (Royal Soc. Open Sci. 2025) - Technology review with policy and safety considerations\n\n#### Policy & Strategic Perspectives\n- [Artificial Intelligence for Science](https:\u002F\u002Fwww.csiro.au\u002F-\u002Fmedia\u002Fd61\u002Fai4science-report\u002Fai-for-science-report-2022.pdf) (CSIRO 2022) - Landmark report analyzing AI adoption across 98% of scientific fields over 60 years\n- [AI for Science 2025](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fd42473-025-00161-3) (Fudan University & Nature 2025) - Comprehensive report on AI's transformative impact across 7 scientific fields, 28 research directions, and 90+ challenges\n- [AI in science evidence review](https:\u002F\u002Fscientificadvice.eu\u002Fscientific-outputs\u002Fai-in-science-evidence-review-report\u002F) (European Scientific Advice 2024) - Policy-focused evidence review on AI's impact in research\n\n### 🚀 AI Scientist & Autonomous Research (2024-2025 Breakthroughs)\n- [The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.06292) (2024.08) - First fully autonomous research system\n- [The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.08066) (2025.04) - Enhanced autonomous research with agentic tree search\n- [AI-Researcher: Autonomous Scientific Innovation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.18705) (2025.05) - Autonomous research pipeline from literature to publication with Scientist-Bench evaluation framework\n- [InternAgent: When Agent Becomes the Scientist -- Building Closed-Loop System from Hypothesis to Verification](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.16938) (2025.05) - Multi-agent system achieving #1 on MLE-Bench with closed-loop research automation\n- [Autonomous Scientific Discovery Through Hierarchical AI Scientist Systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.15951) (2025.07) - Self-evolving multi-agent research systems\n- [ChemCrow: Augmenting large-language models with chemistry tools](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.05376) (2023.04) - LLM agents for chemistry research\n- [Autonomous chemical research with large language models](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-023-06792-0) - Automated chemical experimentation\n- [Coscientist: Autonomously planning and executing scientific experiments](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-023-06792-1) - Robotic lab automation\n\n### Recent Advances & Domain Applications\n- [AlphaFold: Protein Structure Prediction](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-021-03819-2)\n- [AI for Materials Discovery](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41578-023-00540-6) \n- [Large Language Models in Chemistry](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.05852) (2024.02)\n- [Cell2Sentence: Teaching Large Language Models the Language of Biology](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.06147) (ICML 2024) - LLMs for single-cell transcriptomics\n- [Scaling Large Language Models for Next-Generation Single-Cell Analysis](https:\u002F\u002Fwww.biorxiv.org\u002Fcontent\u002F10.1101\u002F2025.04.14.648850v2) (2025.04) - 27B parameter biological language models\n- [Boltz-1: Democratizing Biomolecular Interaction Modeling](https:\u002F\u002Fwww.biorxiv.org\u002Fcontent\u002F10.1101\u002F2024.11.19.624167v2) (bioRxiv 2024) - First fully open-source model achieving AlphaFold3-level accuracy\n- [MOOSE: Large Language Models for Automated Open-domain Scientific Hypotheses Discovery](https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.02726) (ACL 2024) - First work showing LLMs can generate novel and valid scientific hypotheses, ICML Best Poster Award\n- [Earth-Agent: Unlocking the Full Landscape of Earth Observation with Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2509.23141) (2025.09) - LLM agent framework for Earth Observation with 104 specialized tools and multi-modal analysis\n- [MedAgents: Large Language Models as Collaborators for Zero-shot Medical Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.10537) (ACL 2024) - Multi-disciplinary collaboration framework for medical reasoning using role-playing LLM agents\n- [MedAgentGym: A Scalable Agentic Training Environment for Code-Centric Reasoning in Biomedical Data Science](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.04405) (2025.06) - Specialized training environment for biomedical AI agents with code-centric reasoning\n- [Paper2Web: Let's Make Your Paper Alive!](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.15842) (2025.10) - AI-powered transformation of academic papers into interactive websites with comprehensive evaluation framework\n- [DeepAnalyze: Agentic Large Language Models for Autonomous Data Science](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.16872) (2025.10) - First agentic LLM for autonomous data science with curriculum-based training\n- [Democratizing AI scientists using ToolUniverse](https:\u002F\u002Farxiv.org\u002Fabs\u002F2509.23426) (2025.09) - Universal ecosystem for building AI scientists from any LLM with 600+ scientific tools\n- [TxAgent: An AI Agent for Therapeutic Reasoning Across a Universe of Tools](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.10970) (2025.03) - AI agent achieving 92.1% accuracy in drug reasoning, outperforming GPT-4o by 25.8%\n- [Aviary: Training Language Agents on Challenging Scientific Tasks](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.21154) (2024.12) - Language agent training framework for scientific discovery\n- [Galactica: A Large Language Model for Science](https:\u002F\u002Farxiv.org\u002Fabs\u002F2211.09085) (2022.11)\n\n### 📈 Evaluation & Benchmarking\n- [ScienceAgentBench (ICLR 2025)](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002FScienceAgentBench) - 102 executable tasks from 44 peer-reviewed papers across 4 disciplines with containerized evaluation\n- [Scientist-Bench](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FAI-Researcher) - Comprehensive benchmark for comparing LLM Agent-generated research outcomes with high-quality scientific work\n- [SciTrust: Evaluating the Trustworthiness of Large Language Models for Science](https:\u002F\u002Fimpact.ornl.gov\u002Fen\u002Fpublications\u002Fscitrust-evaluating-the-trustworthiness-of-large-language-models-) (2024) - Scientific LLM trustworthiness evaluation framework\n- [SciBench: Evaluating College-Level Scientific Problem-Solving Abilities](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.10635) (2023) - Scientific reasoning benchmarks\n- [ChartCoder Evaluation](https:\u002F\u002Faclanthology.org\u002F2025.acl-long.363\u002F) - Chart-to-code generation benchmarks\n\n---\n\n## 🔬 Domain-Specific Applications\n\n### 🧬 Biology & Medicine\n\n#### Protein & Drug Discovery\n- [AlphaFold](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold) - Protein structure prediction\n- [ColabFold (2025 Updates)](https:\u002F\u002Fgithub.com\u002Fsokrypton\u002FColabFold) - AlphaFold\u002FESMFold accessible implementation with AF3 JSON export, database updates\n- [OpenFold3](https:\u002F\u002Fgithub.com\u002Faqlaboratory\u002Fopenfold-3) - Fully open-source (Apache 2.0) biomolecular structure prediction reproducing AlphaFold3, free for academic and commercial use (Columbia AlQuraishi Lab & OpenFold Consortium, 2025)\n- [Protenix](https:\u002F\u002Fgithub.com\u002Fbytedance\u002FProtenix) - Trainable PyTorch reproduction of AlphaFold 3\n- [Chai-1](https:\u002F\u002Fgithub.com\u002Fchaidiscovery\u002Fchai-lab) - Multi-modal foundation model for biomolecular structure prediction (proteins, small molecules, DNA, RNA, glycans) achieving SOTA across benchmarks, with optional MSA\u002Ftemplate support (Chai Discovery, 2024)\n- [Boltz](https:\u002F\u002Fgithub.com\u002Fjwohlwend\u002Fboltz) - First fully open-source model achieving AlphaFold3-level accuracy with 1000x faster binding affinity prediction (MIT)\n- [BoltzGen](https:\u002F\u002Farxiv.org\u002Fabs\u002F2511.18345) - De novo protein binder design via generative model, achieving nanomolar binding for 66% of novel targets tested (MIT, 2025)\n- [xfold](https:\u002F\u002Fgithub.com\u002FShenggan\u002Fxfold) - Democratizing AlphaFold3: PyTorch reimplementation to accelerate protein structure prediction research\n- [MegaFold](https:\u002F\u002Fgithub.com\u002FSupercomputing-System-AI-Lab\u002FMegaFold\u002F) - Cross-platform system optimizations for accelerating AlphaFold3 training with 1.73x speedup and 1.23x memory reduction\n- [Graphormer](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FGraphormer) - General-purpose deep learning backbone for molecular modeling\n- [DiffDock](https:\u002F\u002Fgithub.com\u002Fgcorso\u002FDiffDock) - Diffusion-based molecular docking achieving SOTA blind docking performance, treating ligand pose prediction as generative diffusion over SE(3), with DiffDock-L update for improved generalization (MIT CSAIL, ICLR 2023)\n- [targetdiff](https:\u002F\u002Fgithub.com\u002Fguanjq\u002Ftargetdiff) - 3D Equivariant Diffusion for Target-Aware Molecule Generation (ICLR2023)\n- [ReQFlow](https:\u002F\u002Fgithub.com\u002FAngxiaoYue\u002FReQFlow) - Rectified Quaternion Flow for efficient protein backbone generation, 37× faster than RFDiffusion with 0.972 designability (ICML 2025)\n- [BioEmu](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fbioemu) - Microsoft's generative model for sampling protein equilibrium conformations 100,000× faster than MD simulations, predicting domain motions, local unfolding and cryptic binding pockets on a single GPU (Science 2025)\n- [ProteinMPNN](https:\u002F\u002Fgithub.com\u002Fdauparas\u002FProteinMPNN) - Deep learning-based protein sequence design (inverse folding) from backbone structures, achieving 52.4% sequence recovery vs 32.9% for Rosetta, core tool in modern protein design pipelines (Baker Lab, Science 2022)\n- [RFdiffusion3](https:\u002F\u002Fgithub.com\u002FRosettaCommons\u002FRFdiffusion) - Latest RFdiffusion for protein structure design with 10× speedup and atom-level precision (December 2025)\n- [IgGM](https:\u002F\u002Fgithub.com\u002FTencentAI4S\u002FIgGM) - Generative foundation model for functional antibody and nanobody design, supporting de novo generation, affinity maturation, inverse design, structure prediction, and humanization (Tencent AI4S, ICLR 2025)\n- [DrugAssist](https:\u002F\u002Fgithub.com\u002Fblazerye\u002FDrugAssist) - LLM-based molecular optimization tool\n- [mint](https:\u002F\u002Fgithub.com\u002FVarunUllanat\u002Fmint) - Learning the language of protein-protein interactions\n- [Mol-Instructions](https:\u002F\u002Fgithub.com\u002Fzjunlp\u002FMol-Instructions) - Large-scale biomolecular instruction dataset for chemistry\u002Fbiology LLMs (ICLR2024)\n- [Uni-Mol](https:\u002F\u002Fgithub.com\u002Fdeepmodeling\u002FUni-Mol) - Universal 3D molecular pretraining framework with 209M conformations, scaling to 1.1B parameters (Uni-Mol2) on 800M conformations for molecular property prediction, docking, and quantum chemistry (ICLR 2023, NeurIPS 2024)\n- [ChemBERTa](https:\u002F\u002Fgithub.com\u002Fseyonechithrananda\u002Fbert-loves-chemistry) - Chemical language model\n- [DeepChem](https:\u002F\u002Fgithub.com\u002Fdeepchem\u002Fdeepchem) - Machine learning for chemistry\n- [DeepMol](https:\u002F\u002Fgithub.com\u002FBioSystemsUM\u002FDeepMol) - Unified ML\u002FDL framework for drug discovery workflows, integrating RDKit, DeepChem, and scikit-learn with SHAP explainability\n- [RDKit](https:\u002F\u002Fgithub.com\u002Frdkit\u002Frdkit) - Cheminformatics toolkit\n- [ESM3](https:\u002F\u002Fgithub.com\u002Fevolutionaryscale\u002Fesm) - 98B-parameter frontier generative model jointly reasoning over protein sequence, structure, and function, trained on 2.78 billion proteins; generated a novel fluorescent protein (esmGFP) with only 58% sequence identity to known GFPs (EvolutionaryScale, 2024)\n- [ESMFold](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fesm) - Protein structure prediction from ESM models\n\n#### Genomics & Bioinformatics\n- [RhoFold+](https:\u002F\u002Fgithub.com\u002Fml4bio\u002FRhoFold) - End-to-end RNA 3D structure prediction using RNA language model pretrained on 23.7M sequences, outperforming existing methods and human expert groups on RNA-Puzzles and CASP15 (Nature Methods 2024)\n- [Evo 2](https:\u002F\u002Fgithub.com\u002FArcInstitute\u002Fevo2) - Arc Institute's 40B-parameter genome foundation model trained on 9 trillion nucleotides from all domains of life, supporting 1M base pair context for generalist DNA\u002FRNA\u002Fprotein prediction and design (Nature 2026)\n- [LucaOne](https:\u002F\u002Fgithub.com\u002FLucaOne\u002FLucaOne) - Generalized biological foundation model with unified nucleic acid and protein language, integrating DNA\u002FRNA\u002Fprotein sequences (Nature Machine Intelligence 2025)\n- [Geneformer](https:\u002F\u002Fgithub.com\u002Flcrawlab\u002FGeneformer) - Single-cell transformer foundation model pretrained on 104M human transcriptomes via masked gene prediction, enabling transfer learning for cell type classification, gene network analysis, and in silico perturbation with limited labeled data (Nature 2023, V2 2024)\n- [scFoundation](https:\u002F\u002Fgithub.com\u002Fbiomap-research\u002FscFoundation) - 100M-parameter foundation model pretrained on 50M+ human single-cell transcriptomes covering ~20,000 genes, achieving SOTA on gene expression enhancement, drug response and perturbation prediction (Nature Methods 2024)\n- [scGPT](https:\u002F\u002Fgithub.com\u002Fbowang-lab\u002FscGPT) - Single-cell analysis with transformers\n- [Cell2Sentence](https:\u002F\u002Fgithub.com\u002Fvandijklab\u002Fcell2sentence) - Teaching Large Language Models the Language of Biology through single-cell transcriptomics (ICML 2024)\n- [ChatSpatial](https:\u002F\u002Fgithub.com\u002Fcafferychen777\u002FChatSpatial) - MCP server enabling spatial transcriptomics analysis via natural language, integrating 60+ methods including SpaGCN, Cell2location, LIANA+, CellRank for Visium, Xenium, MERFISH platforms\n- [Enformer](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fdeepmind-research\u002Ftree\u002Fmaster\u002Fenformer) - Gene expression prediction\n- [DNABERT](https:\u002F\u002Fgithub.com\u002Fjerryji1993\u002FDNABERT) - DNA sequence analysis\n- [scBERT](https:\u002F\u002Fgithub.com\u002FTencentAILabHealthcare\u002FscBERT) - Single-cell BERT for gene expression\n- [GenePT](https:\u002F\u002Fgithub.com\u002Fyiqunchen\u002FGenePT) - Generative pre-training for genomics\n- [DNA Claude Analysis](https:\u002F\u002Fgithub.com\u002Fshmlkv\u002Fdna-claude-analysis) - Interactive personal genome analysis toolkit using Claude Code and Python. Parses raw genotyping data from consumer DNA services and analyzes SNPs across 17 categories including health risks, pharmacogenomics, ancestry, and nutrition, with a terminal-style HTML dashboard.\n\n#### Medical AI & Clinical Applications\n- [MedSAM](https:\u002F\u002Fgithub.com\u002Fbowang-lab\u002FMedSAM) - Universal medical image segmentation foundation model trained on 1.57M image-mask pairs across 10 imaging modalities and 30+ cancer types, with MedSAM2 extending to 3D and video segmentation (Nature Communications 2024)\n- [MedAgents](https:\u002F\u002Fgithub.com\u002Fgersteinlab\u002FMedAgents) - Multi-disciplinary collaboration framework for zero-shot medical reasoning using role-playing LLM agents (ACL 2024)\n- [MedAgentGym](https:\u002F\u002Fgithub.com\u002Fwshi83\u002FMedAgentGym) - Scalable agentic training environment for code-centric reasoning in biomedical data science\n\n### ⚛ Chemistry & Materials\n\n#### LLM for Chemistry\n- [LLM4Chemistry](https:\u002F\u002Fgithub.com\u002FOpenDFM\u002FLLM4Chemistry) - Curated paper list about LLMs for chemistry covering fine-tuning, reasoning, multi-modal models, agents, and benchmarks (COLING 2025)\n\n#### Materials Discovery\n- [GNoME](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmaterials_discovery) - DeepMind's graph neural network for materials exploration, discovering 2.2M new crystal structures (380K most stable) equivalent to 800 years of traditional research, with 520K+ materials dataset open-sourced (Nature 2023)\n- [FAIRChem (OMat24)](https:\u002F\u002Fgithub.com\u002FFAIR-Chem\u002Ffairchem) - Meta's comprehensive ML ecosystem for materials\u002Fchemistry with 118M+ DFT calculations, EquiformerV2 models achieving top Matbench Discovery performance\n- [MACE](https:\u002F\u002Fgithub.com\u002FACEsuit\u002Fmace) - Machine learning interatomic potentials\n- [CHGNet](https:\u002F\u002Fgithub.com\u002FCederGroupHub\u002Fchgnet) - Universal pretrained neural network potential with charge and magnetic moment awareness, trained on 1.5M+ Materials Project inorganic structures for charge-informed molecular dynamics and phase diagram prediction (Berkeley, Nature Machine Intelligence 2023 Cover)\n- [MatterSim](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmattersim) - Deep learning atomistic model across elements, temperatures, and pressures\n- [Crystal Graph CNNs](https:\u002F\u002Fgithub.com\u002Ftxie-93\u002Fcgcnn) - Crystal property prediction\n- [MatBench](https:\u002F\u002Fgithub.com\u002Fmaterialsproject\u002Fmatbench) - Materials informatics benchmark\n- [Best of Atomistic Machine Learning](https:\u002F\u002Fgithub.com\u002FJuDFTteam\u002Fbest-of-atomistic-machine-learning) - Curated list of atomistic ML projects for materials science\n\n#### Chemical Synthesis\n- [AiZynthFinder](https:\u002F\u002Fgithub.com\u002FMolecularAI\u002Faizynthfinder) - AstraZeneca's industrial-grade retrosynthetic planning tool using MCTS to recursively decompose molecules into purchasable precursors, with multi-step route scoring and support for custom one-step models (v4.0, 2024)\n- [Molecular Transformers](https:\u002F\u002Fgithub.com\u002Fpschwllr\u002FMolecularTransformer) - AI for chemical reaction prediction and synthesis planning\n\n### 🌌 Physics & Astronomy\n\n#### Machine Learning for Physics\n- [FermiNet](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fferminet) - DeepMind's neural network for ab-initio quantum chemistry, directly solving the many-electron Schrödinger equation via variational Monte Carlo with antisymmetric wavefunctions, extended to excited states (Phys. Rev. Research 2020, Science 2024)\n- [JAX-MD](https:\u002F\u002Fgithub.com\u002Fjax-md\u002Fjax-md) - Molecular dynamics in JAX\n- [Neural ODEs](https:\u002F\u002Fgithub.com\u002Frtqichen\u002Ftorchdiffeq) - Differential equations with neural networks\n- [Physics-Informed Neural Networks](https:\u002F\u002Fgithub.com\u002Fmaziarraissi\u002FPINNs) - Physics-constrained ML\n- [EquiformerV2](https:\u002F\u002Fgithub.com\u002Fatomicarchitects\u002Fequiformer_v2) - Improved equivariant Transformer for 3D atomic graphs (ICLR2024)\n- [Equiformer](https:\u002F\u002Fgithub.com\u002Fatomicarchitects\u002Fequiformer) - Equivariant graph attention Transformer (ICLR2023)\n\n#### Astronomy & Astrophysics\n- [AstroPy](https:\u002F\u002Fgithub.com\u002Fastropy\u002Fastropy) - Python astronomy tools\n- [Gaia Archive](https:\u002F\u002Fgea.esac.esa.int\u002Farchive\u002F) - Stellar data for ML\n- [DeepSphere](https:\u002F\u002Fgithub.com\u002Fdeepsphere\u002Fdeepsphere-pytorch) - Spherical CNNs for astronomy\n\n### 🌍 Earth & Climate Science\n\n#### Climate Modeling\n- [GenCast](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fgraphcast) - Google DeepMind's diffusion-based ensemble weather forecasting model at 0.25° resolution, outperforming ECMWF ENS on 97.2% of targets up to 15 days ahead, with open-source code and weights (Nature 2024)\n- [Aurora](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Faurora) - Microsoft's foundation model for the Earth system supporting weather, air pollution, and ocean wave forecasting at multiple resolutions, trained on 1M+ hours of diverse atmospheric data (Nature 2025)\n- [ClimaX](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FClimaX) - First foundation model for weather and climate by Microsoft, Vision Transformer-based architecture trained on heterogeneous datasets (ICML 2023)\n- [NeuralGCM](https:\u002F\u002Fgithub.com\u002Fneuralgcm\u002Fneuralgcm) - Google Research's hybrid ML\u002Fphysics atmospheric model combining learned dynamics with physical constraints, outperforming traditional models on 2-15 day forecasts and 40-year climate simulation, developed with ECMWF (Nature 2024)\n- [NVIDIA Earth-2](https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fearth2studio) - World's first fully open, accelerated weather AI software stack with Medium Range forecasting and Nowcasting models using generative AI (January 2026)\n- [Pangu-Weather](https:\u002F\u002Fgithub.com\u002F198808xc\u002FPangu-Weather) - Huawei's 3D high-resolution global weather forecast model at 0.25° resolution, first AI method to comprehensively outperform traditional NWP across all variables and lead times, integrated into ECMWF operational forecasts (Nature 2023)\n- [Prithvi WxC](https:\u002F\u002Fhuggingface.co\u002Fibm\u002Fprithvi-wxc) - IBM-NASA open-source 2.3B parameter weather and climate foundation model trained on 160 MERRA-2 variables, runs on desktop with fine-tuned variants for climate downscaling and gravity wave parameterization\n- [ClimateBench](https:\u002F\u002Fgithub.com\u002Fduncanwp\u002FClimateBench) - Climate data benchmark for ML models\n- [WeatherBench](https:\u002F\u002Fgithub.com\u002Fpangeo-data\u002FWeatherBench) - Weather prediction benchmark\n- [WeatherGFT](https:\u002F\u002Fgithub.com\u002Fblack-yt\u002FWeatherGFT) - Physics-AI hybrid modeling for fine-grained weather forecasting (NeurIPS'24)\n- [Awesome Large Weather Models](https:\u002F\u002Fgithub.com\u002Fjaychempan\u002FAwesome-LWMs) - Curated list of large weather models for AI Earth science\n- [TerraTorch](https:\u002F\u002Fgithub.com\u002FIBM\u002Fterratorch) - Python toolkit for fine-tuning geospatial foundation models\n- [Earth-Agent](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FEarth-Agent) - LLM agent framework for Earth Observation with 104 specialized tools across 5 functional kits\n- [AI for Earth](https:\u002F\u002Fplanetarycomputer.microsoft.com\u002F) - Microsoft's environmental AI\n\n### 🌾 Agriculture & Ecology\n\n#### Agricultural AI\n- [PlantNet](https:\u002F\u002Fplantnet.org\u002F) - Plant identification using AI and citizen science\n- [AgML](https:\u002F\u002Fgithub.com\u002FProject-AgML\u002FAgML) - Agricultural machine learning platform\n\n#### Ecological Modeling\n- [BioSimulators](https:\u002F\u002Fgithub.com\u002Fbiosimulators\u002FBiosimulators) - Biological simulation tools\n- [EcoNet](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FEcoNet) - Ecological modeling and conservation AI\n\n---\n\n## 🤖 Foundation Models for Science\n\n### General Science Models\n- [Galactica](https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai) - Large language model for science\n- [Llemma](https:\u002F\u002Fgithub.com\u002FEleutherAI\u002Fmath-lm) - Open language model for mathematics (7B\u002F34B) trained on Proof-Pile-2, outperforming Minerva at equal scale on MATH benchmark, with tool use and formal theorem proving in Lean without finetuning (EleutherAI, ICLR 2024)\n- [MinervaAI](https:\u002F\u002Fgithub.com\u002Fgoogle-research\u002Fminerva) - Mathematical reasoning\n- [PaLM-2](https:\u002F\u002Fai.google\u002Fdiscover\u002Fpalm2) - Scientific reasoning capabilities\n\n### Domain-Specific Models\n- [ESM](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fesm) - Protein language models\n- [BioNeMo Framework](https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fbionemo-framework) - NVIDIA's open-source platform for building and adapting biological AI models at scale, bundling ESM-2, Geneformer, MolMIM and DNA embedding models with recipes for single-GPU to multi-node training (2025)\n- [ChemGPT](https:\u002F\u002Fhuggingface.co\u002Fncfrey\u002FChemGPT-1.2B) - Chemistry-focused language model\n- [BioGPT](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FBioGPT) - Biomedical text generation\n\n---\n\n## 📈 Datasets & Benchmarks\n\n### Multidisciplinary\n- [Hugging Face Datasets](https:\u002F\u002Fhuggingface.co\u002Fdatasets) - Comprehensive ML research datasets and scientific data collections\n- [Google Dataset Search](https:\u002F\u002Fdatasetsearch.research.google.com\u002F) - Find scientific datasets\n\n### Biology & Medicine\n- [Protein Data Bank](https:\u002F\u002Fwww.rcsb.org\u002F) - Protein structures\n- [ChEMBL](https:\u002F\u002Fwww.ebi.ac.uk\u002Fchembl\u002F) - Chemical bioactivity data\n- [Human Protein Atlas](https:\u002F\u002Fwww.proteinatlas.org\u002F) - Protein expression data\n- [Chinese Medical Dataset](https:\u002F\u002Fgithub.com\u002FMengqi97\u002Fchinese-medical-dataset) - Comprehensive collection of Chinese medical datasets for AI research\n\n### Chemistry & Materials\n- [Materials Project](https:\u002F\u002Fnext-gen.materialsproject.org\u002F) - Computational materials database\n- [QM9](https:\u002F\u002Fquantum-machine.org\u002Fdatasets\u002F) - Small molecule properties\n- [Open Catalyst Project](https:\u002F\u002Fopencatalystproject.org\u002F) - Catalyst discovery\n\n### Physics\n- [LIGO Open Science Center](https:\u002F\u002Fgwosc.org\u002F) - Gravitational wave data\n- [Particle Data Group](https:\u002F\u002Fpdg.lbl.gov\u002F) - Particle physics data\n- [OpenQuantumMaterials](https:\u002F\u002Fwww.quantum-materials.org\u002F) - Quantum materials data\n\n---\n\n## 💻 Computing Frameworks\n\n### Machine Learning\n- [PyTorch](https:\u002F\u002Fpytorch.org\u002F) - Deep learning framework\n- [JAX](https:\u002F\u002Fgithub.com\u002Fjax-ml\u002Fjax) - High-performance ML research\n- [TensorFlow](https:\u002F\u002Ftensorflow.org\u002F) - End-to-end ML platform\n\n### Scientific Computing\n- [NumPy](https:\u002F\u002Fnumpy.org\u002F) - Numerical computing\n- [SciPy](https:\u002F\u002Fscipy.org\u002F) - Scientific computing\n- [Scikit-learn](https:\u002F\u002Fscikit-learn.org\u002F) - Machine learning library\n\n### Scientific Machine Learning Frameworks\n- [SciML](https:\u002F\u002Fsciml.ai\u002F) - Scientific machine learning ecosystem\n- [DifferentialEquations.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FDifferentialEquations.jl) - Multi-language suite for high-performance differential equation solving and scientific machine learning (3.0k+ stars)\n- [ModelingToolkit.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FModelingToolkit.jl) - Acausal modeling framework for automatically parallelized scientific machine learning (1.5k+ stars)\n- [SciMLBenchmarks.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FSciMLBenchmarks.jl) - Scientific machine learning benchmarks & differential equation solvers\n- [NeuralPDE.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FNeuralPDE.jl) - Physics-informed neural networks (PINNs) for solving partial differential equations (1.1k+ stars)\n- [DiffEqFlux.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FDiffEqFlux.jl) - Neural ordinary differential equations with O(1) backprop and GPU support (900+ stars)\n- [Optimization.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FOptimization.jl) - Unified interface for local, global, gradient-based and derivative-free optimization (800+ stars)\n- [PaddleScience](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleScience) - SDK & library for AI-driven scientific computing applications\n- [Flux.jl](https:\u002F\u002Fgithub.com\u002FFluxML\u002FFlux.jl) - Machine learning in Julia\n\n### Specialized Frameworks\n- [MDAnalysis](https:\u002F\u002Fgithub.com\u002FMDAnalysis\u002Fmdanalysis) - Molecular dynamics analysis\n- [MDtrajNet](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.16301) - Neural network foundation model that directly generates MD trajectories bypassing force calculations, accelerating simulations by up to 100× with equivariant Transformer architecture (2025)\n- [ASE](https:\u002F\u002Fwiki.fysik.dtu.dk\u002Fase\u002F) - Atomic Simulation Environment for materials modeling\n- [PyMC](https:\u002F\u002Fgithub.com\u002Fpymc-devs\u002Fpymc) - Probabilistic programming\n- [OpenMM](https:\u002F\u002Fgithub.com\u002Fopenmm\u002Fopenmm) - High-performance molecular simulation toolkit\n\n---\n\n## 🎓 Educational Resources\n\n### Courses & Tutorials\n- [AI for Everyone (Coursera)](https:\u002F\u002Fwww.coursera.org\u002Flearn\u002Fai-for-everyone) - Basic AI concepts\n- [CS229 Machine Learning](https:\u002F\u002Fcs229.stanford.edu\u002F) - Stanford ML course\n- [MIT 6.034 Artificial Intelligence](https:\u002F\u002Focw.mit.edu\u002Fcourses\u002F6-034-artificial-intelligence-fall-2010\u002F) - AI fundamentals\n\n### Open Access Educational Materials\n- [SciML Book](https:\u002F\u002Fgithub.com\u002FSciML\u002FSciMLBook) - Parallel Computing and Scientific Machine Learning: MIT 18.337J\u002F6.338J course materials (1.9k+ stars)\n- [Dive into Deep Learning](https:\u002F\u002Fd2l.ai\u002F) - Interactive deep learning book with code implementations\n- [The Elements of Statistical Learning](https:\u002F\u002Fhastie.su.stanford.edu\u002FElemStatLearn\u002F) - Classic ML textbook freely available\n- [Neural Networks and Deep Learning](http:\u002F\u002Fneuralnetworksanddeeplearning.com\u002F) - Free online book by Michael Nielsen\n\n### 📋 Paper Collections & Repositories\n- [Awesome Scientific Language Models](https:\u002F\u002Fgithub.com\u002Fyuzhimanhua\u002FAwesome-Scientific-Language-Models) - Curated scientific LLM papers (260+ models)\n- [Awesome LLM Scientific Discovery](https:\u002F\u002Fgithub.com\u002FHKUST-KnowComp\u002FAwesome-LLM-Scientific-Discovery) - LLM papers for scientific discovery\n- [AI4Research Papers](https:\u002F\u002Fgithub.com\u002Fdu-nlp-lab\u002FLLM4SR) - LLM for scientific research papers\n- [Physics-Informed Neural Networks Papers](https:\u002F\u002Fgithub.com\u002Fidrl-lab\u002FPINNpapers) - PINN research collection\n- [Scientific Computing with ML Papers](https:\u002F\u002Fsciml.ai\u002Fpapers\u002F) - Scientific ML paper repository\n- [Simulation-Based Inference Papers & Tools](https:\u002F\u002Fsimulation-based-inference.org\u002Fpapers\u002F) - Community-maintained SBI research portal with papers and software\n- [Awesome AI Scientist Papers](https:\u002F\u002Fgithub.com\u002Fopenags\u002FAwesome-AI-Scientist-Papers) - Autonomous AI scientist research\n- [Awesome Agents for Science](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002Fawesome-agents4science) - LLM agents across scientific domains\n\n### YouTube Channels\n- [Two Minute Papers](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FKárolyZsolnai) - AI research summaries\n- [3Blue1Brown](https:\u002F\u002Fwww.youtube.com\u002Fc\u002F3blue1brown) - Mathematical concepts\n- [AI Coffee Break](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FAICoffeeBreak) - AI paper reviews\n- [Steve Brunton](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FEigensteve) - Data-driven methods\n- [Nathan Kutz](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FNathanKutz) - Applied mathematics\n- [Physics Informed Machine Learning](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FPIML) - SciML tutorials\n\n---\n\n## 🏛 Research Communities\n\n### Conferences\n- [NeurIPS](https:\u002F\u002Fneurips.cc\u002F) - Machine learning conference\n- [ICML](https:\u002F\u002Ficml.cc\u002F) - International Conference on Machine Learning\n- [AI for Science Workshop](https:\u002F\u002Fai4sciencecommunity.github.io\u002F) - Specialized workshops\n\n### Organizations\n- [Partnership on AI](https:\u002F\u002Fpartnershiponai.org\u002F) - AI research collaboration\n- [Allen Institute for AI](https:\u002F\u002Fallenai.org\u002F) - AI research institute\n- [OpenAI](https:\u002F\u002Fopenai.com\u002F) - AI research and deployment\n\n### Online Communities\n- [r\u002FMachineLearning](https:\u002F\u002Freddit.com\u002Fr\u002FMachineLearning) - ML discussions\n- [AI Alignment Forum](https:\u002F\u002Fwww.alignmentforum.org\u002F) - AI safety research\n- [Distill](https:\u002F\u002Fdistill.pub\u002F) - Visual explanations of ML\n\n---\n\n## 📚 Related Awesome Lists\n\nThis project builds upon and complements several excellent resources:\n\n### 🎯 Specialized Collections\n- [awesome-ai4s](https:\u002F\u002Fgithub.com\u002Fhyperai\u002Fawesome-ai4s) - 200+ AI for Science papers with Chinese interpretations\n- [Awesome AI Scientist Papers](https:\u002F\u002Fgithub.com\u002Fopenags\u002FAwesome-AI-Scientist-Papers) - Autonomous AI scientist research\n- [Awesome Scientific Machine Learning](https:\u002F\u002Fgithub.com\u002FMartinuzziFrancesco\u002Fawesome-scientific-machine-learning) - Physics-informed ML and SciML\n- [Awesome Agents for Science](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002Fawesome-agents4science) - LLM agents across scientific domains\n- [Awesome LLM Agents Scientific Discovery](https:\u002F\u002Fgithub.com\u002Fzhoujieli\u002FAwesome-LLM-Agents-Scientific-Discovery) - Biomedical AI agents\n- [Awesome Foundation Models for Weather and Climate](https:\u002F\u002Fgithub.com\u002Fshengchaochen82\u002FAwesome-Foundation-Models-for-Weather-and-Climate) - Comprehensive survey of foundation models for weather and climate data understanding\n\n### 📊 Paper & Research Collections\n- [Scientific LLM Papers](https:\u002F\u002Fgithub.com\u002Fyuzhimanhua\u002FAwesome-Scientific-Language-Models) - 260+ scientific language models\n- [LLM4SR Repository](https:\u002F\u002Fgithub.com\u002Fdu-nlp-lab\u002FLLM4SR) - LLM for scientific research survey materials\n- [PINNs Paper Collection](https:\u002F\u002Fgithub.com\u002Fidrl-lab\u002FPINNpapers) - Physics-informed neural networks research\n- [SciML Papers](https:\u002F\u002Fsciml.ai\u002Fpapers\u002F) - Scientific computing and machine learning papers\n\n### 🌟 Key Insights from These Collections\n- **Current Focus**: Shift from tool-level assistance to autonomous scientific agents\n- **Emerging Trends**: Multi-modal scientific models, self-improving research systems\n- **Research Gaps**: Evaluation frameworks, ethical governance, human-AI collaboration\n- **Future Directions**: Fully autonomous discovery cycles, robotic lab integration\n\n---\n\n## 🤝 Contributing\n\nWe welcome contributions! Please see our [Contributing Guidelines](CONTRIBUTING.md) for details.\n\n### How to Contribute\n1. Fork this repository\n2. Add your resource in the appropriate section\n3. Ensure the format matches existing entries\n4. Submit a pull request with a clear description\n\n### Contribution Guidelines\n- Ensure the resource is actively maintained\n- Include a brief, clear description\n- Check for duplicates before adding\n- Use proper markdown formatting\n\n---\n\n## 📄 License\n\nThis project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.\n\n---\n\n## 🙏 Acknowledgments\n\nSpecial thanks to all researchers and developers pushing the boundaries of AI for Science. This list is inspired by the awesome community and the transformative potential of AI in scientific discovery.\n\n**Star ⭐ this repository if you find it helpful!**\n\n---\n\n*Last updated: January 2026 - Enhanced with 2025-2026 breakthroughs in autonomous research, equation discovery, and scientific foundation models*","\u003Cdiv align=\"center\">\n  \u003Ch1>✨ 科学领域的超棒AI（AI4Science）✨\u003C\u002Fh1>\n  \n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fai-boost_awesome-ai-for-science_readme_9bb893d60e49.jpg\" alt=\"科学领域的超棒AI横幅\" width=\"100%\">\n  \n  \u003Cp align=\"center\">\n    一份精心整理的清单，汇集了能够加速跨学科\u003Cstrong>科学发现\u003C\u002Fstrong>的优秀AI工具、库、论文、数据集和框架。\n  \u003C\u002Fp>\n  \n  \u003C!-- 保留这些链接。翻译会随README自动更新。 -->\n  \u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fde\u002Fai-boost\u002Fawesome-ai-for-science\">Deutsch\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fen\u002Fai-boost\u002Fawesome-ai-for-science\">English\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fes\u002Fai-boost\u002Fawesome-ai-for-science\">Español\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Ffr\u002Fai-boost\u002Fawesome-ai-for-science\">français\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fja\u002Fai-boost\u002Fawesome-ai-for-science\">日本語\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fko\u002Fai-boost\u002Fawesome-ai-for-science\">한국어\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fpt\u002Fai-boost\u002Fawesome-ai-for-science\">Português\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fru\u002Fai-boost\u002Fawesome-ai-for-science\">Русский\u003C\u002Fa> | \n    \u003Ca href=\"https:\u002F\u002Fzdoc.app\u002Fzh\u002Fai-boost\u002Fawesome-ai-for-science\">中文\u003C\u002Fa>\n  \u003C\u002Fp>\n  \n  \u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fawesome.re\">\u003Cimg src=\"https:\u002F\u002Fawesome.re\u002Fbadge.svg\" alt=\"Awesome\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicense\u002FMIT\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow.svg\" alt=\"许可证：MIT\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fai-boost\u002Fawesome-ai-for-science\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fai-boost\u002Fawesome-ai-for-science.svg?style=social&label=Star\" alt=\"GitHub星标\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fai-boost\u002Fawesome-ai-for-science\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fforks\u002Fai-boost\u002Fawesome-ai-for-science.svg?style=social&label=Fork\" alt=\"GitHub叉子\">\u003C\u002Fa>\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n> AI正彻底改变科学研究——从药物研发、材料设计到气候建模和天体物理学。本仓库收集了最优质的资源，帮助研究人员在其工作中充分利用AI。\n\n## 📚 目录\n\n- [🧪 研究用AI工具](#-ai-tools-for-research)\n- [📄 论文→海报\u002F幻灯片\u002F图文摘要](#-paperposter--slides--graphical-abstract)\n- [📊 图表理解与生成](#-chart-understanding--generation)\n- [🔄 论文转代码与可重复性](#-paper-to-code--reproducibility)\n- [📋 科学文档处理与解析](#-scientific-documentation--parsing)\n- [🧰 研究工作台与插件](#-research-workbench--plugins)\n- [🕸 知识抽取与学术知识图谱](#-knowledge-extraction--scholarly-kgs)\n- [🤖 研究代理与自动化工作流](#-research-agents--autonomous-workflows)\n- [🏷 数据标注与整理](#-data-labeling--curation)\n- [⚗ 科学机器学习](#-scientific-machine-learning)\n- [📖 论文与综述](#-papers--reviews)\n- [🔬 领域特定应用](#-domain-specific-applications)\n  - [🧬 生物学与医学](#-biology--medicine)\n  - [⚛ 化学与材料](#-chemistry--materials)  \n  - [🌌 物理学与天文学](#-physics--astronomy)\n  - [🌍 地球与气候科学](#-earth--climate-science)\n  - [🌾 农业与生态学](#-agriculture--ecology)\n- [🤖 科学领域基础模型](#-foundation-models-for-science)\n- [📈 数据集与基准测试](#-datasets--benchmarks)\n- [💻 计算框架](#-computing-frameworks)\n- [🎓 教育资源](#-educational-resources)\n- [🏛 研究社区](#-research-communities)\n- [📚 相关的Awesome列表](#-related-awesome-lists)\n\n---\n\n## 🧪 研究用AI工具\n\n### 文献与知识管理\n- [Semantic Scholar](https:\u002F\u002Fwww.semanticscholar.org\u002F) - 基于AI的学术搜索引擎（Allen AI）\n- [arXiv](https:\u002F\u002Farxiv.org\u002F) - 开放获取的电子预印本和后印本库\n- [OpenAlex](https:\u002F\u002Fopenalex.org\u002F) - 学术论文和作者的开放目录\n- [CORE](https:\u002F\u002Fcore.ac.uk\u002F) - 开放获取研究论文聚合平台\n\n### 数据分析与可视化\n- [PandasAI](https:\u002F\u002Fgithub.com\u002FSinaptik-AI\u002Fpandas-ai) - 使用自然语言进行对话式数据分析\n- [DeepAnalyze](https:\u002F\u002Fgithub.com\u002Fruc-datalab\u002FDeepAnalyze) - 首个用于自主数据科学的代理型大模型，提供从数据到分析师级报告的端到端流程\n- [AutoViz](https:\u002F\u002Fgithub.com\u002FAutoViML\u002FAutoViz) - 自动化数据可视化，只需极少代码\n- [Chat2Plot](https:\u002F\u002Fgithub.com\u002Fnyanp\u002Fchat2plot) - 通过标准化图表规范实现安全的文本到可视化转换\n\n### 数据标注与注释\n- [Label Studio](https:\u002F\u002Fgithub.com\u002Fheartexlabs\u002Flabel-studio) - 多类型数据标注与注释工具\n- [Snorkel](https:\u002F\u002Fgithub.com\u002Fsnorkel-team\u002Fsnorkel) - 程序化数据标注与弱监督方法\n\n### 研究工作台与插件\n- [Claude Scientific Skills](https:\u002F\u002Fgithub.com\u002FK-Dense-AI\u002Fclaude-scientific-skills) - 为Claude AI提供的全面集合，包含超过125个即用型科学技能模块，覆盖生物信息学、化学信息学、临床研究、机器学习和材料科学等领域\n\n---\n\n## 📄 论文→海报\u002F幻灯片\u002F图文摘要\n\n### 海报生成\n- [Paper2Poster](https:\u002F\u002Fgithub.com\u002FPaper2Poster\u002FPaper2Poster) - 采用解析器-规划器-绘图器架构的多智能体系统，可将`paper.pdf`转换为可编辑的`poster.pptx`，性能优于GPT-4o，且所需token数量减少87%\n- [mPLUG-PaperOwl](https:\u002F\u002Fgithub.com\u002FX-PLUG\u002FmPLUG-DocOwl) - 多模态大模型，用于科学图表和示意图的理解与生成\n\n### 幻灯片与演示文稿生成\n- [Auto-Slides](https:\u002F\u002Fauto-slides.github.io\u002F) - 多智能体学术论文到高质量演示文稿系统，支持交互式优化\n- [PPTAgent](https:\u002F\u002Fgithub.com\u002Ficip-cas\u002FPPTAgent) - 不仅能将文本转化为幻灯片，还具备PPTEval多维度评估功能（EMNLP 2025）\n- [paper2slides](https:\u002F\u002Fgithub.com\u002Ftakashiishida\u002Fpaper2slides) - 利用大模型将arXiv论文转化为Beamer幻灯片\n- [PaperToSlides](https:\u002F\u002Fgithub.com\u002Fjxtse\u002FPaperToSlides) - 基于AI的工具，可自动将学术论文（PDF）转换为演示文稿幻灯片\n- [pdf2slides](https:\u002F\u002Fgithub.com\u002Fha0ranyu\u002Fpdf2slides) - 仅需三行代码即可将PDF文件转换为可编辑的幻灯片\n- [SlideDeck AI](https:\u002F\u002Fgithub.com\u002Fbarun-saha\u002Fslide-deck-ai) - 结合生成式AI，根据文档或主题共同创建PowerPoint演示文稿\n- [AI多智能体演示文稿生成器](https:\u002F\u002Fgithub.com\u002FAzure-Samples\u002Fai-multi-agent-presentation-builder) - Azure Semantic Kernel提供的多智能体PPT生成参考\n\n### 视频与媒体生成\n- [Paper2Video](https:\u002F\u002Fgithub.com\u002Fshowlab\u002FPaper2Video) - 首个针对科学论文自动生成视频的基准测试（NeurIPS 2025）\n- [paper2video](https:\u002F\u002Fgithub.com\u002Fmett29\u002Fpaper2video) - 将arXiv上的研究论文转化为引人入胜的演示文稿及适合YouTube发布的视频\n\n### 网站与交互内容生成\n- [Paper2All](https:\u002F\u002Fgithub.com\u002FYuhangChen1\u002FPaper2All) - 基于AI的工作流，将论文转化为交互式网站、海报和多媒体演示文稿，秉持“让您的论文鲜活起来”的理念。\n\n### 图表与可视化生成  \n*注：有关全面的图表理解与代码生成工具，请参阅[📊 图表理解与生成](#-chart-understanding--generation)章节*\n\n---\n\n## 📊 图表理解与生成\n\n### 图表转代码与可复现性\n- [ChartCoder (ACL 2025)](https:\u002F\u002Faclanthology.org\u002F2025.acl-long.363\u002F) - 多模态大语言模型用于图表转代码，7B参数规模的模型性能超越更大规模的开源多模态大语言模型。\n- [ChartAssistant \u002F ChartAst (ACL 2024)](https:\u002F\u002Fgithub.com\u002FOpenGVLab\u002FChartAst) - 通用图表理解和推理模型。\n- [Chart-to-Text 数据集](https:\u002F\u002Fgithub.com\u002Fvis-nlp\u002FChart-to-text) - 大规模图表摘要数据集，用于训练图表描述能力。\n\n### 科学可视化工具\n- [Chat2Plot](https:\u002F\u002Fgithub.com\u002Fnyanp\u002Fchat2plot) - 通过标准化图表规范实现安全的文本到可视化转换。\n- [AutoViz](https:\u002F\u002Fgithub.com\u002FAutoViML\u002FAutoViz) - 自动化数据可视化，只需极少代码。\n- [PlotlyAI](https:\u002F\u002Fplotly.com\u002Fai\u002F) - 基于AI的数据可视化与仪表盘创建工具。\n\n---\n\n## 🔄 论文转代码与可复现性\n\n### 自动化代码生成\n- [AutoP2C](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.20115) - LLM代理框架，可根据学术论文生成可运行的代码仓库。\n- [ResearchCodeAgent](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.20117) - 多智能体系统，用于自动编码研究方法。\n- [ToolMaker](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2502.11705) - 将包含代码的论文转化为可调用的代理工具。\n\n### 实验自动化\n- [BioProBench](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FGreatCaptainNemo\u002FBioProBench) - 综合基准测试，用于自动评估LLM在生物实验方案及程序理解方面的能力。\n- [Alhazen](https:\u002F\u002Fchanzuckerberg.github.io\u002Falhazen\u002F) - 从科学文献中提取实验元数据和操作流程信息。\n\n---\n\n## 📋 科学文档处理与解析\n\n### 高性能文档处理\n- [MinerU (2024\u002F2025)](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FMinerU) - SOTA多模态文档解析模型，拥有12亿参数，性能超越GPT-4o，可将PDF转换为适合LLM使用的Markdown\u002FJSON格式。\n- [PDF-Extract-Kit (2024)](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FPDF-Extract-Kit) - 全面的PDF内容提取工具包，具备布局检测、公式识别和OCR功能。\n- [Docling (IBM, AAAI 2025)](https:\u002F\u002Fresearch.ibm.com\u002Fpublications\u002Fdocling-an-efficient-open-source-toolkit-for-ai-driven-document-conversion) - 支持多种格式（PDF\u002FDOCX\u002FPPTX\u002FHTML\u002F图片）到结构化数据（Markdown\u002FJSON）的转换，并能重建布局、恢复表格和公式。\n- [Nougat (Meta AI)](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fnougat) - 针对学术文档的神经网络光学理解技术，可将科学PDF转换为支持数学公式的Markdown格式。\n- [PaddleOCR 3.0 (2024\u002F2025)](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR) - 先进的OCR技术，结合PP-StructureV3文档解析，准确率提升13%，支持80多种语言。\n- [Unstructured](https:\u002F\u002Fgithub.com\u002FUnstructured-IO\u002Funstructured) - 生产级ETL工具，用于将复杂文档转换为结构化格式，并提供开源API。\n- [Marker](https:\u002F\u002Fgithub.com\u002Fdatalab-to\u002Fmarker) - 高精度PDF→Markdown\u002FJSON\u002FHTML转换工具，尤其擅长处理表格、公式和代码块，并配有基准测试脚本。\n- [S2ORC doc2json (AllenAI)](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fs2orc-doc2json) - 大规模PDF\u002FLaTeX\u002FJATS解析工具，可将数百万篇论文转换为标准化JSON格式。\n- [GROBID](https:\u002F\u002Fgithub.com\u002Fkermitt2\u002Fgrobid) - 用于从学术文档中提取结构化元数据的机器学习软件。\n- [Science-Parse \u002F SPv2 (AllenAI)](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fscience-parse) - 可将科学论文解析为结构化字段（标题\u002F作者\u002F章节\u002F参考文献）。\n\n### 生产级流水线与数据准备\n- [IBM 数据准备工具包：PDF→Parquet](https:\u002F\u002Fibm.github.io\u002Fdata-prep-kit\u002Ftransforms\u002Flanguage\u002Fpdf2parquet\u002F) - 大规模科学文档摄取流水线，配备优化配置。\n- [Mozilla 文档转Markdown](https:\u002F\u002Fgithub.com\u002Fmozilla-ai\u002Fdocument-to-markdown) - 基于Docling的解析工具，提供UI\u002FCLI演示，便于快速原型开发。\n\n### 图表与表格提取\n- [PDFFigures2](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fpdffigures2) - 从学术PDF中提取图表、表格、图注及章节标题。\n- [TableBank](https:\u002F\u002Fgithub.com\u002Fdoc-analysis\u002FTableBank) - 大规模表格检测与识别数据集，附带预训练模型。\n\n### 科学文献RAG与分析\n- [PaperQA2](https:\u002F\u002Fgithub.com\u002Ffuture-house\u002Fpaper-qa) - 高精度RAG系统，适用于科学PDF，支持引用、代理式RAG及矛盾检测。\n- [OpenScholar](https:\u002F\u002Fgithub.com\u002FAkariAsai\u002FOpenScholar) - 基于检索增强的语言模型，能够综合4500万篇论文的科学文献，其引用准确性达到人类专家水平，在ScholarQABench基准测试中比GPT-4o高出5%（Nature 2026，UW & Ai2联合发布）。\n- [paper-reviewer](https:\u002F\u002Fgithub.com\u002Fdeep-diver\u002Fpaper-reviewer) - 可根据arXiv论文生成综合性评论，并将其转化为博客文章。\n\n---\n\n## 🧰 研究工作台与插件\n\n### 交互式研究环境\n- [Jupyter AI (JupyterLab扩展)](https:\u002F\u002Fgithub.com\u002Fjupyterlab\u002Fjupyter-ai) - 官方Jupyter扩展，提供`%%ai`魔法命令和侧边栏聊天助手，可连接多个模型提供商及本地推理引擎。\n- [Notebook Intelligence (NBI)](https:\u002F\u002Fgithub.com\u002Fnotebook-intelligence\u002Fnotebook-intelligence) - JupyterLab中的AI编码助手，具备代理模式，支持任意LLM提供商（2025年及以后）。\n- [Google Colab AI功能](https:\u002F\u002Fcolab.research.google.com\u002F) - 集成AI辅助功能，适用于数据科学和研究笔记本。\n\n### 文献管理插件\n- [PapersGPT for Zotero](https:\u002F\u002Fgithub.com\u002Fpapersgpt\u002Fpapersgpt-for-zotero) - 在Zotero中实现多PDF对话、检索与引用功能，支持商业或本地模型（如Ollama），并兼容MCP协议。\n- [Zotero-GPT (MuiseDestiny)](https:\u002F\u002Fgithub.com\u002FMuiseDestiny\u002Fzotero-gpt) - 经典开源插件，可在Zotero内进行文献问答与摘要生成。\n- [Better BibTeX for Zotero](https:\u002F\u002Fretorque.re\u002Fzotero-better-bibtex\u002F) - 增强的引文键管理与LaTeX集成功能。\n\n### 科学写作与协作\n- [Notion AI](https:\u002F\u002Fwww.notion.so\u002Fproduct\u002Fai) - 基于AI的研究笔记记录与知识管理工具。\n- [Obsidian Smart Connections](https:\u002F\u002Fgithub.com\u002Fbrianpetro\u002Fobsidian-smart-connections) - 基于AI的笔记链接与研究图谱导航功能。\n- [Research Rabbit](https:\u002F\u002Fwww.researchrabbit.ai\u002F) - 基于AI的文献发现与科研网络映射工具。\n\n---\n\n## 🕸 知识抽取与学术知识图谱\n\n### 知识图谱构建\n- [iText2KG](https:\u002F\u002Fgithub.com\u002FAuvaLab\u002Fitext2kg) - 基于大模型的增量式知识图谱构建，包含实体抽取与Neo4j可视化\n- [GraphGen](https:\u002F\u002Fgithub.com\u002Fopen-sciencelab\u002FGraphGen) - 由知识图谱指导的合成数据生成，用于大模型微调，在科学问答（GPQA-Diamond）和数学推理（AIME）任务上表现优异\n- [KoPA](https:\u002F\u002Fgithub.com\u002Fzjukg\u002FKoPA) - 结构感知前缀适配方法，用于将大模型与知识图谱集成（ACM MM 2024）\n- [Scholarly KGQA](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.09841) - 基于大模型的学术知识图谱问答系统（ArXiv论文）\n\n### 知识图谱相关资源\n- [Awesome-LLM-KG](https:\u002F\u002Fgithub.com\u002FRManLuo\u002FAwesome-LLM-KG) - 关于统一大模型与知识图谱的论文综合整理\n\n---\n\n## 🤖 研究代理与自主工作流\n\n### 自主研究系统（2024–2025年突破）\n- [AI科学家v1（2024）](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.06292) - 首个完全自主的研究系统：假设→实验→写作→评审的全流程模拟\n- [AI科学家v2（2025）](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.08066) - 引入智能体树搜索机制，减少对模板的依赖，首次产出研讨会级别接受的论文\n- [DeepScientist](https:\u002F\u002Fgithub.com\u002FResearAI\u002FDeepScientist) - 首个在前沿AI任务中逐步超越人类最先进水平的系统（分别提升183.7%、1.9%、7.9%），历经一个月的自主探索，累计使用超过2万GPU小时\n- [Kosmos](https:\u002F\u002Fgithub.com\u002Fjimmc414\u002FKosmos) - 扩展版自主AI科学家，支持200个并行代理运行，执行4.2万行代码，每次运行分析1500篇论文，准确率达79.4%，并取得7项科学发现（Edison Scientific）\n- [AlphaResearch](https:\u002F\u002Fgithub.com\u002Fanswers111\u002Falpha-research) - 自主算法发现系统，结合进化搜索与同行评审奖励模型，在圆盘堆积问题上达到最佳性能\n- [AI-Researcher](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FAI-Researcher) - 自主化流程，涵盖文献综述→假设提出→算法实现→发表级写作，并通过Scientist-Bench进行评估\n- [Agent Laboratory](https:\u002F\u002Fagentlaboratory.github.io\u002F) - 多智能体工作流，支持完整的科研周期，配合AgentRxiv实现累积性发现\n- [InternAgent](https:\u002F\u002Fgithub.com\u002FAlpha-Innovator\u002FInternAgent) - 封闭式多智能体系统，覆盖12项科学任务的假设提出到验证全过程，在MLE-Bench榜单中排名第一（36.44%）\n- [freephdlabor](https:\u002F\u002Fgithub.com\u002Fltjed\u002Ffreephdlabor) - 首个完全可定制的开源多智能体框架，从创意构思到LaTeX论文，全程自动化管理动态工作流\n- [ToolUniverse](https:\u002F\u002Fgithub.com\u002Fmims-harvard\u002FToolUniverse) - 通过整合600余种科学工具，将任意大模型转化为研究系统，推动AI科学家的普及化（哈佛MIMS）\n- [LabClaw](https:\u002F\u002Fgithub.com\u002Fwu-yc\u002FLabClaw) - 生物医学AI代理技能操作层，涵盖生物学、药理学、医学、数据科学和文献检索等7个领域，共211份生产就绪的SKILL.md文件，支持模块化的实验室推理与实验方案组合，适用于斯坦福LabOS兼容的代理\n- [Robin](https:\u002F\u002Fgithub.com\u002FFuture-House\u002Frobin) - FutureHouse的端到端科学发现多智能体系统，协调文献检索（Crow\u002FFalcon）和数据分析（Finch）代理，首次利用AI生成药物发现，识别出ripasudil作为新型干性AMD治疗药物（2025年）\n- [Aviary](https:\u002F\u002Fgithub.com\u002FFuture-House\u002Faviary) - 面向复杂科学任务的语言代理训练平台，包括DNA操作、文献检索和蛋白质工程等\n- [Curie](https:\u002F\u002Fgithub.com\u002FJust-Curieous\u002FCurie) - 利用AI代理开展自动化且严谨的科学实验\n- [POPPER](https:\u002F\u002Fgithub.com\u002Fsnap-stanford\u002FPOPPER) - 基于智能体的序列式证伪法，实现自动化假设检验\n- [autoresearch](https:\u002F\u002Fgithub.com\u002Fkarpathy\u002Fautoresearch) - Andrej Karpathy的自主大模型研究框架：AI代理在真实训练环境中夜间运行实验，自动编辑代码→5分钟训练→评估，循环往复，单GPU每晚可完成约100次实验\n- [UniScientist](https:\u002F\u002Fgithub.com\u002FUniPat-AI\u002FUniScientist) - 覆盖50余门学科的通用科学研究智能，将大模型定位为跨学科创新引擎，由人类专家负责验证；300亿参数模型在5项研究基准测试中表现优于Claude Opus和GPT\n\n### 评估与基准测试\n- [ScienceAgentBench（ICLR 2025）](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002FScienceAgentBench) - 汇集44篇同行评审论文中的102个可执行任务，覆盖4大学科领域，并采用容器化方式评估\n- [BuildArena](https:\u002F\u002Fgithub.com\u002FAI4Science-WestlakeU\u002FBuildArena) - 首个面向物理规则的交互式基准测试，专为LLM代理设计，可在物理模拟器中建造火箭、汽车和桥梁，并配备3D空间几何库\n- [SciTrust（2024）](https:\u002F\u002Fimpact.ornl.gov\u002Fen\u002Fpublications\u002Fscitrust-evaluating-the-trustworthiness-of-large-language-models-) - 科学类大模型的可信度评估框架（真实性、幻觉、阿谀奉承）\n- [SciCode](https:\u002F\u002Fgithub.com\u002Fscicode-bench\u002FSciCode) - 由科学家精心策划的研究编码基准，包含16个子领域的338个子问题（物理、数学、材料、生物、化学），评估大模型在真实科学编程任务中的表现，并提供黄金标准解决方案（NeurIPS 2024）\n- [SciBench](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.10635) - 多领域高校级别科学问题解决能力评估\n\n### 学术评审与评价\n- [AgentReview](https:\u002F\u002Fagentreview.github.io\u002F) - 模拟学术同行评审生态系统的LLM代理\n- [LLM-Peer-Review](https:\u002F\u002Fgithub.com\u002FVijayGKR\u002FLLM-Peer-Review) - 基于网页的大模型辅助稿件评审与注释应用\n\n### 领域专用研究代理\n- [Aletheia](https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.10177) - 谷歌DeepMind推出的基于Gemini Deep Think的自主数学研究代理，无需人工干预即可自主解决700个埃尔德什猜想中的4个开放问题，并生成完整的研究论文（2026年2月）\n- [AlphaGeometry](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphageometry) - DeepMind开发的奥林匹克级别几何定理证明器，结合了神经语言模型与符号推理引擎。AlphaGeometry2以金牌选手水平解决了84%的IMO几何题（42\u002F50）（Nature 2024）\n- [Goedel-Prover-V2](https:\u002F\u002Fgithub.com\u002FGoedel-LM\u002FGoedel-Prover-V2) - Lean 4中最强大的开源自动定理证明器。8B模型在MiniF2F基准上达到84.6%的准确率，与DeepSeek-Prover-V2-671B相当；32B模型通过自校正机制进一步提升至90.4%，采用支架式数据合成和验证器引导的证明精炼技术（普林斯顿大学，2025年）\n- [BioDiscoveryAgent](https:\u002F\u002Fgithub.com\u002Fsnap-stanford\u002FBioDiscoveryAgent) - 用于生物发现与科研自动化的AI代理\n- [MOOSE](https:\u002F\u002Fgithub.com\u002FZonglinY\u002FMOOSE) - 基于大型语言模型的开放式科学假设自动发现系统（ACL 2024，ICML最佳海报奖）\n- [ChemCrow](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.05376) - 集成工具的化学研究LLM代理\n- [Coscientist](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-023-06792-1) - 自主化学实验规划与执行系统\n\n---\n\n## 🏷 数据标注与整理\n\n### 弱监督与自动标注\n- [Snorkel](https:\u002F\u002Fgithub.com\u002Fsnorkel-team\u002Fsnorkel) - 面向科学数据集的程序化数据标注与弱监督方法\n- [PandasAI](https:\u002F\u002Fgithub.com\u002FSinaptik-AI\u002Fpandas-ai) - 使用自然语言进行对话式数据分析与可视化\n\n---\n\n## ⚗ 科学机器学习\n\n### 神经微分方程\n- [torchdiffeq](https:\u002F\u002Fgithub.com\u002Frtqichen\u002Ftorchdiffeq) - PyTorch实现的神经ODE\n- [torchdyn](https:\u002F\u002Fgithub.com\u002FDiffEqML\u002Ftorchdyn) - PyTorch中的神经微分方程\n- [diffrax](https:\u002F\u002Fgithub.com\u002Fpatrick-kidger\u002Fdiffrax) - JAX中用于数值求解微分方程的库\n- [DifferentialEquations.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FDifferentialEquations.jl) - Julia中的微分方程工具包\n- [DiffEqFlux.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FDiffEqFlux.jl) - Julia中的神经微分方程\n\n### 物理信息神经网络\n- [DeepXDE](https:\u002F\u002Fgithub.com\u002Flululxvi\u002Fdeepxde) - 用于求解偏微分方程的深度学习库\n- [Lang-PINN](https:\u002F\u002Fopenreview.net\u002Fforum?id=ONEyVpgK34) - 基于LLM的多智能体系统，能够根据自然语言任务描述构建可训练的PINN模型，使均方误差降低3–5个数量级，执行成功率提升50%以上（ICLR 2026口头报告）\n- [PINNs](https:\u002F\u002Fgithub.com\u002Fmaziarraissi\u002FPINNs) - 物理信息神经网络\n- [NVIDIA PhysicsNeMo](https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fphysicsnemo) - 大规模物理驱动机器学习模型的开源框架（原Modulus，2025年更名）\n- [PINA](https:\u002F\u002Fgithub.com\u002FmathLab\u002FPINA) - 用于高级建模的PyTorch版物理信息神经网络\n- [SciANN](https:\u002F\u002Fgithub.com\u002Fsciann\u002Fsciann) - 基于Keras的科学神经网络\n- [NeuralPDE.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FNeuralPDE.jl) - Julia中的物理信息神经网络\n\n### 神经算子与模型发现\n- [DeepONet](https:\u002F\u002Fgithub.com\u002Flululxvi\u002Fdeeponet) - 非线性算子学习\n- [PySINDy](https:\u002F\u002Fgithub.com\u002Fdynamicslab\u002Fpysindy) - 非线性动力系统的稀疏识别\n- [PySR](https:\u002F\u002Fgithub.com\u002FMilesCranmer\u002FPySR) - 高性能符号回归工具，用于从数据中发现可解释的科学方程，采用Python\u002FJulia后端的多种群进化搜索，在物理学和天文学领域广泛应用（剑桥大学，NeurIPS 2023）\n- [LLM-SR](https:\u002F\u002Fgithub.com\u002Fdeep-symbolic-mathematics\u002FLLM-SR) - 利用LLM进行科学方程发现与符号回归，结合代码生成与进化搜索（ICLR 2025口头报告）\n- [Fourier Neural Operator](https:\u002F\u002Fgithub.com\u002Fneuraloperator\u002Fneuraloperator) - 在傅里叶空间中学习算子\n\n---\n\n## 📖 论文与综述\n\n### 基础性论文\n- [机器学习在科学计量分析中的应用](https:\u002F\u002Farxiv.org\u002Fabs\u002F2109.10073)（2021年9月）- 综合性综述\n- [人工智能在科学领域的进展与挑战](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.04346)（2023年3月）- 当前领域现状\n- [面向科学的基础模型](https:\u002F\u002Farxiv.org\u002Fabs\u002F2205.15075)（2022年5月）- 大型模型在科研中的应用\n- [神经常微分方程](https:\u002F\u002Farxiv.org\u002Fabs\u002F1806.07366)（2018年6月）- 神经ODE领域的突破性工作\n- [物理信息神经网络](https:\u002F\u002Farxiv.org\u002Fabs\u002F1711.10561)（2017年11月）- 基于物理约束的深度学习\n- [人工智能时代的科学发现](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-023-06221-2) - Nature关于AI在科学中作用的综述\n\n### 📊 综合性综述与评论（2024—2025年）\n\n#### 人工智能在科学研究中的应用\n- 《人工智能辅助科学发现综述》（2025年2月）——从文献检索到同行评审，全面概述大型语言模型在科学研究生命周期中的作用\n- 《AI4Research：面向科学研究的人工智能综述》（2025年7月）——系统性地梳理人工智能在科研领域的分类体系\n- 《量子、原子尺度及连续介质系统中的科学人工智能》（2023年7月）——由63位作者共同撰写的跨科学尺度的综合性技术综述\n- 《从自动化到自主性：大型语言模型在科学发现中的应用综述》（2025年5月）——提出工具、分析师和科学家三个层次的分类框架\n- 《从“AI for Science”到“Agentic Science”：自主科学发现综述》（2025年8月）——涵盖生命科学、化学、材料科学和物理学等领域的自主科学发现综合评述\n- 《用于科学发现的代理型人工智能：进展、挑战与未来方向》（2025年3月）——全面回顾人工智能代理在科学领域的应用\n- 《迈向科学智能：基于大型语言模型的科学代理综述》（2025年3月）——科学人工智能代理系统\n\n#### 科学领域大型语言模型\n- 《科学领域大型语言模型及其应用综合综述》（2024年6月）——覆盖多个学科的260余种科学专用大型语言模型\n- 《科学领域大型语言模型综述：从数据基础到代理前沿》（2025年8月）——以数据为中心视角审视科学领域大型语言模型\n- 《科学领域大型语言模型：生物与化学领域综述》（2024年1月）——聚焦特定领域的科学专用大型语言模型\n\n#### 科学机器学习\n- 《基于物理信息神经网络的科学机器学习：现状与展望》（2022年1月）——全面回顾物理信息神经网络\n- 《物理信息神经网络及其扩展》（2024年8月）——最新进展及变体介绍\n- 《基于模拟的推断前沿》（PNAS，2020年）——Cranmer等人关于SBI在科学计算中应用的基础性综述\n- 《从理论到实践：科学计算中神经算子的实用入门》（2025年3月）——聚焦DeepONet、FNO和PCANet等实现方法的指南\n- 《神经算子的架构、变体与性能：比较综述》（2025年）——系统分析DeepONets、积分核算子以及基于Transformer的神经算子\n- 《环境科学的基础模型综述》（2025年4月）——环境领域的应用\n- 《生物信息学中的基础模型》（2025年）——生物学领域的基础模型\n- 《材料发现的基础模型》（2025年）——关于材料人工智能的展望\n\n#### 不确定性量化\n- 《科学机器学习中的不确定性量化：方法、指标与比较》（2023年，J. Comput. Phys.）——Psaros等人提出的PINNs和神经算子中UQ的综合框架\n- 《深度学习不确定性量化方法综述》（2023年）——从不确定性来源角度系统性梳理UQ方法\n\n#### 自动化与无人实验室\n- 《面向化学与材料科学的无人实验室》（2024年，Chem. Rev.）——长达100页的综合性综述，涵盖SDL技术、应用及基础设施\n- 《自主“无人”实验室：技术与政策影响综述》（2025年，Royal Soc. Open Sci.）——结合技术和安全考量的技术综述\n\n#### 政策与战略视角\n- 《面向科学的人工智能》（2022年，CSIRO）——里程碑式报告，分析过去60年间98%科学领域中人工智能的应用情况\n- 《2025年人工智能与科学》（2025年，复旦大学与Nature联合发布）——全面报告人工智能对7个科学领域、28个研究方向以及90余项挑战的变革性影响\n- 《科学领域人工智能证据综述》（2024年，欧洲科学咨询委员会）——聚焦政策的人工智能对科研影响的证据性评估\n\n### 🚀 AI科学家与自主研究（2024—2025年突破）\n- 《AI科学家：迈向完全自动化的开放式科学发现》（2024年8月）——首个完全自主的研究系统\n- 《AI科学家-v2：基于代理树搜索的车间级自动化科学发现》（2025年4月）——通过代理树搜索进一步提升自主研究能力\n- 《AI研究员：自主科学创新》（2025年5月）——构建从文献检索到论文发表的自主研究流程，并引入Scientist-Bench评估框架\n- 《InternAgent：当代理成为科学家——构建从假设到验证的闭环系统》（2025年5月）——多智能体系统在MLE-Bench榜单中排名第一，实现了闭环研究自动化\n- 《基于层级式AI科学家系统的自主科学发现》（2025年7月）——自进化多智能体研究系统\n- 《ChemCrow：将大型语言模型与化学工具相结合》（2023年4月）——用于化学研究的LLM代理\n- 《利用大型语言模型进行自主化学研究》（2023年，Nature）——自动化化学实验\n- 《Coscientist：自主规划并执行科学实验》（2023年，Nature）——机器人实验室自动化\n\n### 最近进展与领域应用\n- [AlphaFold：蛋白质结构预测](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-021-03819-2)\n- [用于材料发现的AI](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41578-023-00540-6) \n- [化学中的大型语言模型](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.05852)（2024年2月）\n- [Cell2Sentence：教大型语言模型生物学语言](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.06147)（ICML 2024）——用于单细胞转录组学的LLMs\n- [扩展大型语言模型以进行下一代单细胞分析](https:\u002F\u002Fwww.biorxiv.org\u002Fcontent\u002F10.1101\u002F2025.04.14.648850v2)（2025年4月）——270亿参数的生物语言模型\n- [Boltz-1：民主化生物分子相互作用建模](https:\u002F\u002Fwww.biorxiv.org\u002Fcontent\u002F10.1101\u002F2024.11.19.624167v2)（bioRxiv 2024）——首个达到AlphaFold3级别精度的完全开源模型\n- [MOOSE：用于自动化开放域科学假设发现的大型语言模型](https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.02726)（ACL 2024）——首次证明LLMs能够生成新颖且有效的科学假设，荣获ICML最佳海报奖\n- [Earth-Agent：用智能体解锁地球观测的全貌](https:\u002F\u002Farxiv.org\u002Fabs\u002F2509.23141)（2025年9月）——具备104种专业工具和多模态分析能力的地球观测LLM智能体框架\n- [MedAgents：作为零样本医学推理协作者的大型语言模型](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.10537)（ACL 2024）——利用角色扮演LLM智能体的多学科协作医学推理框架\n- [MedAgentGym：面向生物医学数据科学中以代码为中心推理的可扩展智能体训练环境](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.04405)（2025年6月）——专为生物医学AI智能体设计的以代码为中心的推理训练环境\n- [Paper2Web：让您的论文鲜活起来！](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.15842)（2025年10月）——由AI驱动的学术论文交互式网站转化，并配有全面评估框架\n- [DeepAnalyze：用于自主数据科学的智能体型大型语言模型](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.16872)（2025年10月）——首个采用课程式训练的自主数据科学智能体LLM\n- [使用ToolUniverse民主化AI科学家](https:\u002F\u002Farxiv.org\u002Fabs\u002F2509.23426)（2025年9月）——一个通用生态系统，可基于任何LLM构建AI科学家，包含600余种科学工具\n- [TxAgent：跨工具宇宙的治疗推理AI智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.10970)（2025年3月）——在药物推理任务中达到92.1%准确率，比GPT-4o高出25.8%\n- [Aviary：针对挑战性科学任务的语言智能体训练框架](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.21154)（2024年12月）——用于科学发现的语言智能体训练框架\n- [Galactica：面向科学的大型语言模型](https:\u002F\u002Farxiv.org\u002Fabs\u002F2211.09085)（2022年11月）\n\n### 📈 评估与基准测试\n- [ScienceAgentBench（ICLR 2025）](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002FScienceAgentBench) —— 包含来自4个学科44篇同行评审论文的102项可执行任务，并采用容器化评估方式\n- [Scientist-Bench](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FAI-Researcher) —— 用于比较LLM智能体生成的研究成果与高质量科学工作的综合基准\n- [SciTrust：评估大型语言模型在科学领域的可信度](https:\u002F\u002Fimpact.ornl.gov\u002Fen\u002Fpublications\u002Fscitrust-evaluating-the-trustworthiness-of-large-language-models-)（2024年）——科学LLM可信度评估框架\n- [SciBench：评估大学水平的科学问题解决能力](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.10635)（2023年）——科学推理基准测试\n- [ChartCoder评估](https:\u002F\u002Faclanthology.org\u002F2025.acl-long.363\u002F) —— 图表到代码生成基准\n\n---\n\n## 🔬 领域特定应用\n\n### 🧬 生物学与医学\n\n#### 蛋白质与药物发现\n- [AlphaFold](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Falphafold) - 蛋白质结构预测\n- [ColabFold (2025年更新)](https:\u002F\u002Fgithub.com\u002Fsokrypton\u002FColabFold) - 可访问的AlphaFold\u002FESMFold实现，支持AF3 JSON导出及数据库更新\n- [OpenFold3](https:\u002F\u002Fgithub.com\u002Faqlaboratory\u002Fopenfold-3) - 完全开源（Apache 2.0），复现AlphaFold3的生物分子结构预测模型，面向学术与商业用途免费开放（哥伦比亚大学AlQuraishi实验室及OpenFold联盟，2025年）\n- [Protenix](https:\u002F\u002Fgithub.com\u002Fbytedance\u002FProtenix) - 可训练的PyTorch版AlphaFold3复现模型\n- [Chai-1](https:\u002F\u002Fgithub.com\u002Fchaidiscovery\u002Fchai-lab) - 多模态基础模型，用于生物分子结构预测（蛋白质、小分子、DNA、RNA、糖链），在各类基准测试中达到最先进水平，并可选MSA\u002F模板支持（Chai Discovery，2024年）\n- [Boltz](https:\u002F\u002Fgithub.com\u002Fjwohlwend\u002Fboltz) - 首个完全开源、达到AlphaFold3级别精度且结合亲和力预测速度提升1000倍的模型（MIT）\n- [BoltzGen](https:\u002F\u002Farxiv.org\u002Fabs\u002F2511.18345) - 基于生成模型的从头设计蛋白质结合剂，在66%的新靶点测试中实现了纳摩尔级结合能力（MIT，2025年）\n- [xfold](https:\u002F\u002Fgithub.com\u002FShenggan\u002Fxfold) - 普及AlphaFold3：基于PyTorch的重实现，加速蛋白质结构预测研究\n- [MegaFold](https:\u002F\u002Fgithub.com\u002FSupercomputing-System-AI-Lab\u002FMegaFold\u002F) - 跨平台系统优化，使AlphaFold3训练提速1.73倍、内存占用降低1.23倍\n- [Graphormer](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FGraphormer) - 通用深度学习骨干网络，适用于分子建模\n- [DiffDock](https:\u002F\u002Fgithub.com\u002Fgcorso\u002FDiffDock) - 基于扩散的分子对接技术，实现最先进的盲对接性能，将配体构象预测视为SE(3)空间上的生成式扩散过程，并推出DiffDock-L版本以提升泛化能力（MIT CSAIL，ICLR 2023）\n- [targetdiff](https:\u002F\u002Fgithub.com\u002Fguanjq\u002Ftargetdiff) - 面向目标的3D等变扩散模型，用于分子生成（ICLR2023）\n- [ReQFlow](https:\u002F\u002Fgithub.com\u002FAngxiaoYue\u002FReQFlow) - 校正四元数流算法，高效生成蛋白质主链，速度比RFDiffusion快37倍，设计性达0.972（ICML 2025）\n- [BioEmu](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fbioemu) - 微软开发的生成模型，可在单张GPU上以比分子动力学模拟快10万倍的速度采样蛋白质平衡构象，预测结构域运动、局部展开及隐匿结合位点（Science，2025年）\n- [ProteinMPNN](https:\u002F\u002Fgithub.com\u002Fdauparas\u002FProteinMPNN) - 基于深度学习的蛋白质序列设计（逆折叠）方法，直接由主链结构生成序列，序列恢复率可达52.4%，而Rosetta仅为32.9%，是现代蛋白质设计流程中的核心工具（Baker实验室，Science，2022年）\n- [RFdiffusion3](https:\u002F\u002Fgithub.com\u002FRosettaCommons\u002FRFdiffusion) - 最新版本的RFdiffusion，用于蛋白质结构设计，速度提升10倍且精度达到原子级别（2025年12月）\n- [IgGM](https:\u002F\u002Fgithub.com\u002FTencentAI4S\u002FIgGM) - 生成式基础模型，用于功能性抗体及纳米抗体的设计，支持从头生成、亲和力成熟、逆向设计、结构预测及人源化改造（腾讯AI4S，ICLR 2025）\n- [DrugAssist](https:\u002F\u002Fgithub.com\u002Fblazerye\u002FDrugAssist) - 基于大语言模型的分子优化工具\n- [mint](https:\u002F\u002Fgithub.com\u002FVarunUllanat\u002Fmint) - 学习蛋白质-蛋白质相互作用的语言\n- [Mol-Instructions](https:\u002F\u002Fgithub.com\u002Fzjunlp\u002FMol-Instructions) - 面向化学\u002F生物学领域大语言模型的大规模生物分子指令数据集（ICLR2024）\n- [Uni-Mol](https:\u002F\u002Fgithub.com\u002Fdeepmodeling\u002FUni-Mol) - 通用3D分子预训练框架，包含2.09亿种构象；扩展至11亿参数的Uni-Mol2版本则基于8亿种构象，用于分子性质预测、分子对接及量子化学计算（ICLR 2023，NeurIPS 2024）\n- [ChemBERTa](https:\u002F\u002Fgithub.com\u002Fseyonechithrananda\u002Fbert-loves-chemistry) - 化学语言模型\n- [DeepChem](https:\u002F\u002Fgithub.com\u002Fdeepchem\u002Fdeepchem) - 面向化学领域的机器学习工具\n- [DeepMol](https:\u002F\u002Fgithub.com\u002FBioSystemsUM\u002FDeepMol) - 面向药物研发工作流的统一ML\u002FDL框架，集成RDKit、DeepChem和scikit-learn，并提供SHAP解释性\n- [RDKit](https:\u002F\u002Fgithub.com\u002Frdkit\u002Frdkit) - 化学信息学工具包\n- [ESM3](https:\u002F\u002Fgithub.com\u002Fevolutionaryscale\u002Fesm) - 980亿参数的前沿生成模型，可同时推理蛋白质序列、结构与功能，基于27.8亿条蛋白质数据进行训练；该模型还生成了一种新型荧光蛋白（esmGFP），其序列与已知GFP仅58%一致（EvolutionaryScale，2024年）\n- [ESMFold](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fesm) - 基于ESM模型的蛋白质结构预测\n\n#### 基因组学与生物信息学\n- [RhoFold+] (https:\u002F\u002Fgithub.com\u002Fml4bio\u002FRhoFold) - 基于2370万条序列预训练的RNA语言模型实现端到端的RNA三维结构预测，在RNA-Puzzles和CASP15比赛中性能超越现有方法及人类专家团队（Nature Methods 2024）\n- [Evo 2] (https:\u002F\u002Fgithub.com\u002FArcInstitute\u002Fevo2) - Arc研究所基于来自所有生命域的9万亿个核苷酸训练的400亿参数基因组基础模型，支持100万碱基对上下文，用于通用DNA\u002FRNA\u002F蛋白质预测与设计（Nature 2026）\n- [LucaOne] (https:\u002F\u002Fgithub.com\u002FLucaOne\u002FLucaOne) - 具有统一核酸与蛋白质语言的广义生物基础模型，整合DNA\u002FRNA\u002F蛋白质序列（Nature Machine Intelligence 2025）\n- [Geneformer] (https:\u002F\u002Fgithub.com\u002Flcrawlab\u002FGeneformer) - 单细胞Transformer基础模型，通过掩码基因预测任务在1.04亿个人类转录组上预训练，可在少量标注数据下实现迁移学习，用于细胞类型分类、基因网络分析及体外扰动研究（Nature 2023，V2 2024）\n- [scFoundation] (https:\u002F\u002Fgithub.com\u002Fbiomap-research\u002FscFoundation) - 1亿参数的基础模型，在覆盖约2万个基因的5000多万个人类单细胞转录组上预训练，在基因表达增强、药物反应及扰动预测方面达到SOTA水平（Nature Methods 2024）\n- [scGPT] (https:\u002F\u002Fgithub.com\u002Fbowang-lab\u002FscGPT) - 基于Transformer的单细胞分析工具\n- [Cell2Sentence] (https:\u002F\u002Fgithub.com\u002Fvandijklab\u002Fcell2sentence) - 通过单细胞转录组学向大型语言模型教授生物学语言（ICML 2024）\n- [ChatSpatial] (https:\u002F\u002Fgithub.com\u002Fcafferychen777\u002FChatSpatial) - MCP服务器，可通过自然语言进行空间转录组学分析，集成包括SpaGCN、Cell2location、LIANA+、CellRank在内的60余种方法，适用于Visium、Xenium、MERFISH等平台\n- [Enformer] (https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fdeepmind-research\u002Ftree\u002Fmaster\u002Fenformer) - 基因表达预测\n- [DNABERT] (https:\u002F\u002Fgithub.com\u002Fjerryji1993\u002FDNABERT) - DNA序列分析\n- [scBERT] (https:\u002F\u002Fgithub.com\u002FTencentAILabHealthcare\u002FscBERT) - 面向基因表达的单细胞BERT模型\n- [GenePT] (https:\u002F\u002Fgithub.com\u002Fyiqunchen\u002FGenePT) - 面向基因组学的生成式预训练模型\n- [DNA Claude Analysis] (https:\u002F\u002Fgithub.com\u002Fshmlkv\u002Fdna-claude-analysis) - 使用Claude Code和Python构建的交互式个人基因组分析工具包。可解析消费者DNA服务提供的原始基因分型数据，并按健康风险、药物基因组学、祖先溯源、营养等17个类别分析SNP位点，配备终端风格的HTML仪表盘。\n\n#### 医疗AI与临床应用\n- [MedSAM] (https:\u002F\u002Fgithub.com\u002Fbowang-lab\u002FMedSAM) - 通用医学图像分割基础模型，在涵盖10种成像模态和30多种癌症类型的157万张图像-掩码对上训练；MedSAM2进一步扩展至3D及视频分割领域（Nature Communications 2024）\n- [MedAgents] (https:\u002F\u002Fgithub.com\u002Fgersteinlab\u002FMedAgents) - 基于角色扮演LLM代理的零样本医学推理多学科协作框架（ACL 2024）\n- [MedAgentGym] (https:\u002F\u002Fgithub.com\u002Fwshi83\u002FMedAgentGym) - 面向生物医学数据科学中以代码为中心的推理的可扩展代理式训练环境\n\n### ⚛ 化学与材料科学\n\n#### 面向化学的LLM\n- [LLM4Chemistry] (https:\u002F\u002Fgithub.com\u002FOpenDFM\u002FLLM4Chemistry) - 关于化学领域LLM的精选论文列表，涵盖微调、推理、多模态模型、智能体及基准测试等内容（COLING 2025）\n\n#### 材料发现\n- [GNoME] (https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmaterials_discovery) - DeepMind开发的图神经网络用于材料探索，发现了220万个新晶体结构（其中38万个最为稳定），相当于800年传统研究的成果，同时开源了包含52万余种材料的数据集（Nature 2023）\n- [FAIRChem (OMat24)] (https:\u002F\u002Fgithub.com\u002FFAIR-Chem\u002Ffairchem) - Meta构建的综合性材料\u002F化学机器学习生态系统，包含超过1.18亿次DFT计算；EquiformerV2模型在Matbench Discovery榜单中名列前茅\n- [MACE] (https:\u002F\u002Fgithub.com\u002FACEsuit\u002Fmace) - 机器学习原子间势能模型\n- [CHGNet] (https:\u002F\u002Fgithub.com\u002FCederGroupHub\u002Fchgnet) - 具有电荷与磁矩感知能力的通用预训练神经网络势能模型，基于150多万个Materials Project无机结构训练而成，可用于电荷敏感分子动力学模拟及相图预测（伯克利，Nature Machine Intelligence 2023封面文章）\n- [MatterSim] (https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmattersim) - 跨元素、温度和压力的深度学习原子级模型\n- [Crystal Graph CNNs] (https:\u002F\u002Fgithub.com\u002Ftxie-93\u002Fcgcnn) - 晶体性质预测\n- [MatBench] (https:\u002F\u002Fgithub.com\u002Fmaterialsproject\u002Fmatbench) - 材料信息学基准测试\n- [Best of Atomistic Machine Learning] (https:\u002F\u002Fgithub.com\u002FJuDFTteam\u002Fbest-of-atomistic-machine-learning) - 材料科学领域原子级机器学习项目的精选列表\n\n#### 化学合成\n- [AiZynthFinder] (https:\u002F\u002Fgithub.com\u002FMolecularAI\u002Faizynthfinder) - 阿斯利康工业级逆向合成规划工具，采用MCTS算法递归分解分子为可采购的前体，具备多步路线评分功能，并支持自定义单步模型（v4.0，2024）\n- [Molecular Transformers] (https:\u002F\u002Fgithub.com\u002Fpschwllr\u002FMolecularTransformer) - 用于化学反应预测与合成规划的人工智能工具\n\n### 🌌 物理与天文学\n\n#### 面向物理的机器学习\n- [FermiNet] (https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fferminet) - DeepMind开发的从头算量子化学神经网络，通过变分蒙特卡洛方法结合反对称波函数直接求解多电子薛定谔方程，并已扩展至激发态研究（Phys. Rev. Research 2020，Science 2024）\n- [JAX-MD] (https:\u002F\u002Fgithub.com\u002Fjax-md\u002Fjax-md) - JAX中的分子动力学\n- [Neural ODEs] (https:\u002F\u002Fgithub.com\u002Frtqichen\u002Ftorchdiffeq) - 基于神经网络的微分方程\n- [Physics-Informed Neural Networks] (https:\u002F\u002Fgithub.com\u002Fmaziarraissi\u002FPINNs) - 受物理约束的机器学习模型\n- [EquiformerV2] (https:\u002F\u002Fgithub.com\u002Fatomicarchitects\u002Fequiformer_v2) - 改进的等变Transformer，适用于3D原子图（ICLR2024）\n- [Equiformer] (https:\u002F\u002Fgithub.com\u002Fatomicarchitects\u002Fequiformer) - 等变图注意力Transformer（ICLR2023）\n\n#### 天文学与天体物理学\n- [AstroPy] (https:\u002F\u002Fgithub.com\u002Fastropy\u002Fastropy) - Python天文学工具\n- [Gaia Archive] (https:\u002F\u002Fgea.esac.esa.int\u002Farchive\u002F) - 用于机器学习的恒星数据\n- [DeepSphere] (https:\u002F\u002Fgithub.com\u002Fdeepsphere\u002Fdeepsphere-pytorch) - 用于天文学的球面CNN\n\n### 🌍 地球与气候科学\n\n#### 气候建模\n- [GenCast](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fgraphcast) - Google DeepMind基于扩散模型的集合天气预报模型，分辨率为0.25°，在长达15天的预测中，97.2%的任务上表现优于ECMWF ENS，提供开源代码和权重（Nature 2024）\n- [Aurora](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Faurora) - Microsoft的地球系统基础模型，支持多分辨率的天气、空气污染和海洋波浪预报，基于超过100万小时的多样化大气数据进行训练（Nature 2025）\n- [ClimaX](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FClimaX) - Microsoft首个用于天气和气候的基础模型，采用视觉Transformer架构，基于异构数据集训练（ICML 2023）\n- [NeuralGCM](https:\u002F\u002Fgithub.com\u002Fneuralgcm\u002Fneuralgcm) - Google Research的混合机器学习\u002F物理大气模型，结合学习到的动力学与物理约束，在2至15天的预报以及40年的气候模拟中均优于传统模型，由ECMWF共同开发（Nature 2024）\n- [NVIDIA Earth-2](https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fearth2studio) - 全球首个完全开放、加速型的天气AI软件栈，包含中长期预报和临近预报模型，采用生成式AI技术（2026年1月）\n- [Pangu-Weather](https:\u002F\u002Fgithub.com\u002F198808xc\u002FPangu-Weather) - 华为的3D高分辨率全球天气预报模型，分辨率为0.25°，是首个在所有变量和不同预报时效上全面超越传统数值天气预报的人工智能方法，并已集成到ECMWF的业务预报中（Nature 2023）\n- [Prithvi WxC](https:\u002F\u002Fhuggingface.co\u002Fibm\u002Fprithvi-wxc) - IBM-NASA开源的23亿参数天气与气候基础模型，基于160个MERRA-2变量进行训练，可在桌面端运行，并有针对气候降尺度和重力波参数化的微调版本\n- [ClimateBench](https:\u002F\u002Fgithub.com\u002Fduncanwp\u002FClimateBench) - 面向机器学习模型的气候数据基准\n- [WeatherBench](https:\u002F\u002Fgithub.com\u002Fpangeo-data\u002FWeatherBench) - 天气预测基准\n- [WeatherGFT](https:\u002F\u002Fgithub.com\u002Fblack-yt\u002FWeatherGFT) - 物理与人工智能相结合的细粒度天气预报建模（NeurIPS'24）\n- [Awesome Large Weather Models](https:\u002F\u002Fgithub.com\u002Fjaychempan\u002FAwesome-LWMs) - 人工智能地球科学领域大型天气模型的精选列表\n- [TerraTorch](https:\u002F\u002Fgithub.com\u002FIBM\u002Fterratorch) - 用于微调地理空间基础模型的Python工具包\n- [Earth-Agent](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FEarth-Agent) - 面向地球观测的大语言模型代理框架，包含5个功能套件中的104种专用工具\n- [AI for Earth](https:\u002F\u002Fplanetarycomputer.microsoft.com\u002F) - Microsoft的环境AI项目\n\n### 🌾 农业与生态学\n\n#### 农业AI\n- [PlantNet](https:\u002F\u002Fplantnet.org\u002F) - 利用AI和公民科学进行植物识别\n- [AgML](https:\u002F\u002Fgithub.com\u002FProject-AgML\u002FAgML) - 农业机器学习平台\n\n#### 生态建模\n- [BioSimulators](https:\u002F\u002Fgithub.com\u002Fbiosimulators\u002FBiosimulators) - 生物仿真工具\n- [EcoNet](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FEcoNet) - 生态建模与保护AI\n\n---\n\n## 🤖 科学领域的基础模型\n\n### 通用科学模型\n- [Galactica](https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai) - 面向科学领域的大型语言模型\n- [Llemma](https:\u002F\u002Fgithub.com\u002FEleutherAI\u002Fmath-lm) - 开放式数学语言模型（7B\u002F34B），基于Proof-Pile-2训练，在MATH基准测试中与同等规模的Minerva相比表现更优，无需微调即可实现工具使用和Lean形式化定理证明（EleutherAI, ICLR 2024）\n- [MinervaAI](https:\u002F\u002Fgithub.com\u002Fgoogle-research\u002Fminerva) - 数学推理\n- [PaLM-2](https:\u002F\u002Fai.google\u002Fdiscover\u002Fpalm2) - 科学推理能力\n\n### 领域特定模型\n- [ESM](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fesm) - 蛋白质语言模型\n- [BioNeMo Framework](https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fbionemo-framework) - NVIDIA的开源平台，用于大规模构建和调整生物AI模型，整合了ESM-2、Geneformer、MolMIM和DNA嵌入模型，并提供从单GPU到多节点训练的配方（2025）\n- [ChemGPT](https:\u002F\u002Fhuggingface.co\u002Fncfrey\u002FChemGPT-1.2B) - 面向化学的语言模型\n- [BioGPT](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FBioGPT) - 生物医学文本生成\n\n---\n\n## 📈 数据集与基准测试\n\n### 多学科\n- [Hugging Face Datasets](https:\u002F\u002Fhuggingface.co\u002Fdatasets) - 综合性的机器学习研究数据集和科学数据集合\n- [Google Dataset Search](https:\u002F\u002Fdatasetsearch.research.google.com\u002F) - 查找科学数据集\n\n### 生物与医学\n- [Protein Data Bank](https:\u002F\u002Fwww.rcsb.org\u002F) - 蛋白质结构\n- [ChEMBL](https:\u002F\u002Fwww.ebi.ac.uk\u002Fchembl\u002F) - 化学生物活性数据\n- [Human Protein Atlas](https:\u002F\u002Fwww.proteinatlas.org\u002F) - 蛋白质表达数据\n- [Chinese Medical Dataset](https:\u002F\u002Fgithub.com\u002FMengqi97\u002Fchinese-medical-dataset) - 面向AI研究的中国医学数据集综合收藏\n\n### 化学与材料\n- [Materials Project](https:\u002F\u002Fnext-gen.materialsproject.org\u002F) - 计算材料数据库\n- [QM9](https:\u002F\u002Fquantum-machine.org\u002Fdatasets\u002F) - 小分子性质\n- [Open Catalyst Project](https:\u002F\u002Fopencatalystproject.org\u002F) - 催化剂发现\n\n### 物理\n- [LIGO Open Science Center](https:\u002F\u002Fgwosc.org\u002F) - 引力波数据\n- [Particle Data Group](https:\u002F\u002Fpdg.lbl.gov\u002F) - 粒子物理数据\n- [OpenQuantumMaterials](https:\u002F\u002Fwww.quantum-materials.org\u002F) - 量子材料数据\n\n---\n\n## 💻 计算框架\n\n### 机器学习\n- [PyTorch](https:\u002F\u002Fpytorch.org\u002F) - 深度学习框架\n- [JAX](https:\u002F\u002Fgithub.com\u002Fjax-ml\u002Fjax) - 高性能机器学习研究\n- [TensorFlow](https:\u002F\u002Ftensorflow.org\u002F) - 端到端机器学习平台\n\n### 科学计算\n- [NumPy](https:\u002F\u002Fnumpy.org\u002F) - 数值计算\n- [SciPy](https:\u002F\u002Fscipy.org\u002F) - 科学计算\n- [Scikit-learn](https:\u002F\u002Fscikit-learn.org\u002F) - 机器学习库\n\n### 科学机器学习框架\n- [SciML](https:\u002F\u002Fsciml.ai\u002F) - 科学机器学习生态系统\n- [DifferentialEquations.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FDifferentialEquations.jl) - 多语言高性能微分方程求解与科学机器学习工具包（3,000+ 颗星）\n- [ModelingToolkit.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FModelingToolkit.jl) - 用于自动并行化科学机器学习的非因果建模框架（1,500+ 颗星）\n- [SciMLBenchmarks.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FSciMLBenchmarks.jl) - 科学机器学习基准测试及微分方程求解器\n- [NeuralPDE.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FNeuralPDE.jl) - 用于求解偏微分方程的物理信息神经网络（PINNs）（1,100+ 颗星）\n- [DiffEqFlux.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FDiffEqFlux.jl) - 具有 O(1) 反向传播和 GPU 支持的神经常微分方程（900+ 颗星）\n- [Optimization.jl](https:\u002F\u002Fgithub.com\u002FSciML\u002FOptimization.jl) - 用于局部、全局、基于梯度及无导数优化的统一接口（800+ 颗星）\n- [PaddleScience](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleScience) - 面向 AI 驱动科学计算应用的 SDK 和库\n- [Flux.jl](https:\u002F\u002Fgithub.com\u002FFluxML\u002FFlux.jl) - Julia 中的机器学习\n\n### 专用框架\n- [MDAnalysis](https:\u002F\u002Fgithub.com\u002FMDAnalysis\u002Fmdanalysis) - 分子动力学分析\n- [MDtrajNet](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.16301) - 神经网络基础模型，可直接生成 MD 轨迹而无需计算力场，采用等变 Transformer 架构，将模拟速度提升至原来的 100 倍（2025 年）\n- [ASE](https:\u002F\u002Fwiki.fysik.dtu.dk\u002Fase\u002F) - 用于材料建模的原子模拟环境\n- [PyMC](https:\u002F\u002Fgithub.com\u002Fpymc-devs\u002Fpymc) - 概率编程\n- [OpenMM](https:\u002F\u002Fgithub.com\u002Fopenmm\u002Fopenmm) - 高性能分子模拟工具包\n\n---\n\n## 🎓 教育资源\n\n### 课程与教程\n- [面向所有人的 AI（Coursera）](https:\u002F\u002Fwww.coursera.org\u002Flearn\u002Fai-for-everyone) - 基础 AI 概念\n- [CS229 机器学习](https:\u002F\u002Fcs229.stanford.edu\u002F) - 斯坦福大学机器学习课程\n- [MIT 6.034 人工智能](https:\u002F\u002Focw.mit.edu\u002Fcourses\u002F6-034-artificial-intelligence-fall-2010\u002F) - 人工智能基础\n\n### 开放获取教育资源\n- [SciML 书籍](https:\u002F\u002Fgithub.com\u002FSciML\u002FSciMLBook) - 并行计算与科学机器学习：MIT 18.337J\u002F6.338J 课程资料（1,900+ 颗星）\n- [深入深度学习](https:\u002F\u002Fd2l.ai\u002F) - 包含代码实现的交互式深度学习书籍\n- [统计学习要素](https:\u002F\u002Fhastie.su.stanford.edu\u002FElemStatLearn\u002F) - 经典 ML 教材，免费提供\n- [神经网络与深度学习](http:\u002F\u002Fneuralnetworksanddeeplearning.com\u002F) - 迈克尔·尼尔森编写的免费在线书籍\n\n### 📋 论文集与仓库\n- [优秀科学语言模型](https:\u002F\u002Fgithub.com\u002Fyuzhimanhua\u002FAwesome-Scientific-Language-Models) - 精选科学领域 LLM 论文（260+ 模型）\n- [优秀 LLM 科学发现](https:\u002F\u002Fgithub.com\u002FHKUST-KnowComp\u002FAwesome-LLM-Scientific-Discovery) - 用于科学发现的 LLM 论文\n- [AI4Research 论文](https:\u002F\u002Fgithub.com\u002Fdu-nlp-lab\u002FLLM4SR) - 用于科学研究的 LLM 论文\n- [物理信息神经网络论文](https:\u002F\u002Fgithub.com\u002Fidrl-lab\u002FPINNpapers) - PINN 研究汇编\n- [使用 ML 的科学计算论文](https:\u002F\u002Fsciml.ai\u002Fpapers\u002F) - 科学 ML 论文库\n- [基于模拟的推断论文与工具](https:\u002F\u002Fsimulation-based-inference.org\u002Fpapers\u002F) - 社区维护的 SBI 研究门户，包含论文和软件\n- [优秀 AI 科学家论文](https:\u002F\u002Fgithub.com\u002Fopenags\u002FAwesome-AI-Scientist-Papers) - 自主 AI 科学家研究\n- [科学领域的优秀智能体](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002Fawesome-agents4science) - 覆盖多个科学领域的 LLM 智能体\n\n### YouTube 频道\n- [两分钟论文](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FKárolyZsolnai) - AI 研究摘要\n- [3Blue1Brown](https:\u002F\u002Fwww.youtube.com\u002Fc\u002F3blue1brown) - 数学概念\n- [AI Coffee Break](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FAICoffeeBreak) - AI 论文评论\n- [Steve Brunton](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FEigensteve) - 数据驱动方法\n- [Nathan Kutz](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FNathanKutz) - 应用数学\n- [物理信息机器学习](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FPIML) - SciML 教学视频\n\n---\n\n## 🏛 研究社区\n\n### 会议\n- [NeurIPS](https:\u002F\u002Fneurips.cc\u002F) - 机器学习会议\n- [ICML](https:\u002F\u002Ficml.cc\u002F) - 国际机器学习大会\n- [AI for Science Workshop](https:\u002F\u002Fai4sciencecommunity.github.io\u002F) - 专题研讨会\n\n### 组织\n- [AI 合作伙伴关系](https:\u002F\u002Fpartnershiponai.org\u002F) - AI 研究合作组织\n- [艾伦人工智能研究所](https:\u002F\u002Fallenai.org\u002F) - AI 研究机构\n- [OpenAI](https:\u002F\u002Fopenai.com\u002F) - AI 研究与部署\n\n### 在线社区\n- [r\u002FMachineLearning](https:\u002F\u002Freddit.com\u002Fr\u002FMachineLearning) - 机器学习讨论\n- [AI 对齐论坛](https:\u002F\u002Fwww.alignmentforum.org\u002F) - AI 安全研究\n- [Distill](https:\u002F\u002Fdistill.pub\u002F) - ML 的可视化解释\n\n---\n\n## 📚 相关优秀列表\n\n本项目建立在多个优秀资源的基础上，并对其进行了补充：\n\n### 🎯 专用集合\n- [awesome-ai4s](https:\u002F\u002Fgithub.com\u002Fhyperai\u002Fawesome-ai4s) - 200+ 篇 AI for Science 论文，附中文解读\n- [优秀 AI 科学家论文](https:\u002F\u002Fgithub.com\u002Fopenags\u002FAwesome-AI-Scientist-Papers) - 自主 AI 科学家研究\n- [优秀科学机器学习](https:\u002F\u002Fgithub.com\u002FMartinuzziFrancesco\u002Fawesome-scientific-machine-learning) - 物理信息 ML 和 SciML\n- [科学领域的优秀智能体](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002Fawesome-agents4science) - 覆盖多个科学领域的 LLM 智能体\n- [优秀 LLM 智能体科学发现](https:\u002F\u002Fgithub.com\u002Fzhoujieli\u002FAwesome-LLM-Agents-Scientific-Discovery) - 生物医学领域的 AI 智能体\n- [优秀气象与气候基础模型](https:\u002F\u002Fgithub.com\u002Fshengchaochen82\u002FAwesome-Foundation-Models-for-Weather-and-Climate) - 关于气象与气候数据理解的基础模型综合调研\n\n### 📊 论文与研究汇编\n- [科学 LLM 论文](https:\u002F\u002Fgithub.com\u002Fyuzhimanhua\u002FAwesome-Scientific-Language-Models) - 260+ 科学语言模型\n- [LLM4SR 资料库](https:\u002F\u002Fgithub.com\u002Fdu-nlp-lab\u002FLLM4SR) - 用于科学研究的 LLM 调研资料\n- [PINNs 论文集](https:\u002F\u002Fgithub.com\u002Fidrl-lab\u002FPINNpapers) - 物理信息神经网络研究\n- [SciML 论文](https:\u002F\u002Fsciml.ai\u002Fpapers\u002F) - 科学计算与机器学习论文\n\n### 🌟 这些资源集的关键洞察\n- **当前焦点**：从工具级辅助转向自主科学智能体\n- **新兴趋势**：多模态科学模型、自我改进的研究系统\n- **研究空白**：评估框架、伦理治理、人机协作\n- **未来方向**：完全自主的发现循环、机器人实验室集成\n\n---\n\n## 🤝 贡献方式\n\n我们欢迎各位贡献！详情请参阅我们的[贡献指南](CONTRIBUTING.md)。\n\n### 如何贡献\n1. 克隆本仓库\n2. 将您的资源添加到相应章节\n3. 确保格式与现有条目一致\n4. 提交带有清晰说明的拉取请求\n\n### 贡献规范\n- 请确保资源处于活跃维护状态\n- 提供简明扼要的描述\n- 添加前请先检查是否重复\n- 使用规范的 Markdown 格式\n\n---\n\n## 📄 许可证\n\n本项目采用 MIT 许可证授权——详情请参阅 [LICENSE](LICENSE) 文件。\n\n---\n\n## 🙏 致谢\n\n特别感谢所有推动 AI for Science 边界的科研人员和开发者。这份列表深受社区启发，也体现了 AI 在科学发现中的变革性潜力。\n\n**如果您觉得本资源有帮助，请为本仓库点赞 ⭐！**\n\n---\n\n*最后更新：2026年1月——新增了2025至2026年间在自主研究、方程发现及科学基础模型领域的突破性进展*","# Awesome AI for Science 快速上手指南\n\n`awesome-ai-for-science` 并非单一的独立软件，而是一个精选的开源资源列表，汇集了加速科学发现的 AI 工具、库、论文和数据集。本指南将指导你如何获取该列表，并快速部署其中几个核心的通用工具（如文档解析 MinerU 和数据分析 PandasAI）以开始你的科研工作。\n\n## 环境准备\n\n在开始之前，请确保你的开发环境满足以下基本要求：\n\n*   **操作系统**: Linux (推荐 Ubuntu 20.04+), macOS, 或 Windows (建议配合 WSL2 使用)。\n*   **Python 版本**: Python 3.9 - 3.11 (部分最新工具可能需要 3.12，建议优先使用 3.10)。\n*   **包管理器**: `pip` (建议升级至最新版) 或 `conda`。\n*   **硬件要求**: \n    *   基础文本处理：普通 CPU 即可。\n    *   深度学习模型推理\u002F训练（如 MinerU, ChartCoder）：建议配备 NVIDIA GPU (显存 ≥ 8GB)，并安装对应的 CUDA 驱动。\n*   **前置依赖**: \n    *   Git\n    *   (可选) Docker：用于运行部分容器化部署的工具。\n\n> **国内开发者提示**：建议在安装 Python 依赖时配置清华源或阿里源以加速下载。\n> ```bash\n> pip config set global.index-url https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n> ```\n\n## 安装步骤\n\n由于该仓库包含众多工具，以下提供两种安装方式：获取资源列表本身，以及安装两个最具代表性的高频工具。\n\n### 1. 获取资源列表\n克隆仓库以浏览完整的工具清单和本地文档：\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fai-boost\u002Fawesome-ai-for-science.git\ncd awesome-ai-for-science\n```\n\n### 2. 安装核心工具示例\n\n#### 示例 A: 安装 MinerU (高性能科学文档解析)\nMinerU 是将 PDF 论文转换为 LLM 友好格式（Markdown\u002FJSON）的 SOTA 工具。\n\n```bash\n# 创建虚拟环境\npython -m venv mineru_env\nsource mineru_env\u002Fbin\u002Factivate  # Windows 用户请使用: mineru_env\\Scripts\\activate\n\n# 安装 MinerU (推荐使用国内镜像加速 torch 下载)\npip install magic-pdf[full] -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n\n# 下载模型权重 (首次运行会自动下载，也可手动预下载)\n# 具体模型下载指令参考其官方文档，通常涉及 huggingface 镜像配置\nexport HF_ENDPOINT=https:\u002F\u002Fhf-mirror.com\n```\n\n#### 示例 B: 安装 PandasAI (自然语言数据分析)\n用于通过对话方式分析科学实验数据。\n\n```bash\n# 在当前环境或新环境中安装\npip install pandasai -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n\n# 若需使用本地大模型，还需安装对应依赖，例如 ollama 或 langchain\npip install langchain-community\n```\n\n## 基本使用\n\n以下是两个核心工具的最简使用示例，帮助你快速验证环境。\n\n### 场景一：使用 MinerU 解析科学论文 PDF\n将复杂的科学论文 PDF 转换为结构化的 Markdown，保留公式和表格。\n\n```python\nfrom magic_pdf.data.data_reader_writer import FileBasedDataWriter\nfrom magic_pdf.model.doc_analyze_by_custom_model import doc_analyze\n\n# 设置输入输出路径\npdf_file_path = \"paper.pdf\"\noutput_dir = \".\u002Foutput\"\n\n# 初始化写入器\nimage_writer = FileBasedDataWriter(output_dir)\n\n# 执行解析 (假设已自动下载模型)\n# 注意：实际使用时需根据具体版本调整 API 调用方式\ndoc_analyze(\n    pdf_file_path,\n    image_writer=image_writer,\n    start_page_id=0,\n    end_page_id=None,\n    debug_mode=False\n)\n\nprint(f\"解析完成，结果保存在 {output_dir}\")\n```\n\n### 场景二：使用 PandasAI 分析实验数据\n通过自然语言提问来分析 CSV 格式的实验数据。\n\n```python\nimport pandas as pd\nfrom pandasai import Agent\nfrom pandasai.llm import LocalLLM # 或使用 OpenAI, Ollama 等\n\n# 加载实验数据\ndf = pd.read_csv(\"experiment_results.csv\")\n\n# 配置 LLM (此处以伪代码示例，需替换为实际可用的 LLM 端点)\nllm = LocalLLM(api_base=\"http:\u002F\u002Flocalhost:11434\u002Fv1\", model=\"llama3\")\n\n# 初始化 Agent\nagent = Agent(df, config={\"llm\": llm})\n\n# 自然语言提问\nresponse = agent.chat(\"绘制温度与反应速率的关系图，并计算相关系数\")\nprint(response)\n```\n\n### 探索更多工具\n访问克隆后的仓库目录，查看 `README.md` 文件中的分类列表（如 `AI Tools for Research`, `Paper-to-Code`, `Domain-Specific Applications`），根据具体的科研领域（生物、化学、物理等）选择对应的 GitHub 仓库进行单独安装和使用。","某材料科学实验室的研究团队正致力于利用机器学习加速新型电池电解质的筛选，需要在海量文献中快速定位关键数据并复现前沿算法。\n\n### 没有 awesome-ai-for-science 时\n- **资源检索如大海捞针**：研究人员需在 Google Scholar、arXiv 和各类垂直数据库间反复切换，耗时数周才能拼凑出完整的工具链清单，极易遗漏关键的开源库。\n- **跨领域技术壁垒高**：面对物理学或化学专用的 AI 框架（如科学机器学习库），缺乏针对性的入门指引和教育资源，导致非计算机背景的研究者难以上手。\n- **复现成本高昂**：找不到论文对应的代码实现或标准化数据集，团队不得不从零编写数据处理脚本，严重拖慢了从理论验证到实验测试的进度。\n- **工作流割裂**：文献管理、图表生成、知识提取等环节使用分散的工具，缺乏统一的“研究工作台”概念，协作效率低下且容易出错。\n\n### 使用 awesome-ai-for-science 后\n- **一站式精准导航**：团队直接通过分类目录（如\"⚗️ Chemistry & Materials\"）快速锁定了适用的分子生成模型和专用数据集，将调研时间从数周压缩至几天。\n- **降低跨界门槛**：借助列表中整理的“教育资源”和“基础模型”，团队成员迅速掌握了领域特定的 AI 编程范式，实现了从传统模拟到 AI 驱动发现的平滑过渡。\n- **加速算法落地**：利用\"Paper-to-Code\"板块找到的高可复现代码库，团队直接复用了对应的预处理流程，将新电解质分子的筛选周期缩短了 60%。\n- **构建自动化闭环**：参考“研究智能体”和“工作流插件”推荐，搭建了从文献自动解析到实验数据可视化的自动化流水线，显著提升了团队协作的连贯性。\n\nawesome-ai-for-science 不仅是一份资源清单，更是科研团队打破学科壁垒、将 AI 技术转化为实际科学发现加速器的核心枢纽。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fai-boost_awesome-ai-for-science_9bb893d6.jpg","ai-boost","AwesomeGPTS","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fai-boost_3b7b5a01.png",null,"gpt_boost","https:\u002F\u002Fawesomegpts.vip","https:\u002F\u002Fgithub.com\u002Fai-boost",1434,146,"2026-04-05T08:04:03","MIT",1,"","未说明",{"notes":29,"python":27,"dependencies":30},"该仓库是一个资源列表（Awesome List），汇集了用于科学发现的 AI 工具、库、论文和数据集，本身不是一个可直接运行的单一软件工具。因此没有统一的运行环境需求。具体的环境要求（如操作系统、GPU、Python 版本等）需参考列表中各个独立项目（如 MinerU, Paper2Poster, Nougat 等）的各自文档。",[],[32],"其他",[34,35,36,37,38,39,40],"ai-for-science","ai4s","ai4science","awesome","awesome-list","bioinformatics","scientific-ai",2,"ready","2026-03-27T02:49:30.150509","2026-04-06T05:44:06.991237",[],[],[48,64,73,81,89,97],{"id":49,"name":50,"github_repo":51,"description_zh":52,"stars":53,"difficulty_score":41,"last_commit_at":54,"category_tags":55,"status":42},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。",84991,"2026-04-05T10:45:23",[56,57,58,59,60,32,61,62,63],"图像","数据工具","视频","插件","Agent","语言模型","开发框架","音频",{"id":65,"name":66,"github_repo":67,"description_zh":68,"stars":69,"difficulty_score":70,"last_commit_at":71,"category_tags":72,"status":42},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",[60,56,62,61,32],{"id":74,"name":75,"github_repo":76,"description_zh":77,"stars":78,"difficulty_score":70,"last_commit_at":79,"category_tags":80,"status":42},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。",74913,"2026-04-05T10:44:17",[61,56,62,32],{"id":82,"name":83,"github_repo":84,"description_zh":85,"stars":86,"difficulty_score":25,"last_commit_at":87,"category_tags":88,"status":42},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",[62,32],{"id":90,"name":91,"github_repo":92,"description_zh":93,"stars":94,"difficulty_score":25,"last_commit_at":95,"category_tags":96,"status":42},2234,"scikit-learn","scikit-learn\u002Fscikit-learn","scikit-learn 是一个基于 Python 构建的开源机器学习库，依托于 SciPy、NumPy 等科学计算生态，旨在让机器学习变得简单高效。它提供了一套统一且简洁的接口，涵盖了从数据预处理、特征工程到模型训练、评估及选择的全流程工具，内置了包括线性回归、支持向量机、随机森林、聚类等在内的丰富经典算法。\n\n对于希望快速验证想法或构建原型的数据科学家、研究人员以及 Python 开发者而言，scikit-learn 是不可或缺的基础设施。它有效解决了机器学习入门门槛高、算法实现复杂以及不同模型间调用方式不统一的痛点，让用户无需重复造轮子，只需几行代码即可调用成熟的算法解决分类、回归、聚类等实际问题。\n\n其核心技术亮点在于高度一致的 API 设计风格，所有估算器（Estimator）均遵循相同的调用逻辑，极大地降低了学习成本并提升了代码的可读性与可维护性。此外，它还提供了强大的模型选择与评估工具，如交叉验证和网格搜索，帮助用户系统地优化模型性能。作为一个由全球志愿者共同维护的成熟项目，scikit-learn 以其稳定性、详尽的文档和活跃的社区支持，成为连接理论学习与工业级应用的最",65628,"2026-04-05T10:10:46",[62,32,57],{"id":98,"name":99,"github_repo":100,"description_zh":101,"stars":102,"difficulty_score":41,"last_commit_at":103,"category_tags":104,"status":42},3364,"keras","keras-team\u002Fkeras","Keras 是一个专为人类设计的深度学习框架，旨在让构建和训练神经网络变得简单直观。它解决了开发者在不同深度学习后端之间切换困难、模型开发效率低以及难以兼顾调试便捷性与运行性能的痛点。\n\n无论是刚入门的学生、专注算法的研究人员，还是需要快速落地产品的工程师，都能通过 Keras 轻松上手。它支持计算机视觉、自然语言处理、音频分析及时间序列预测等多种任务。\n\nKeras 3 的核心亮点在于其独特的“多后端”架构。用户只需编写一套代码，即可灵活选择 TensorFlow、JAX、PyTorch 或 OpenVINO 作为底层运行引擎。这一特性不仅保留了 Keras 一贯的高层易用性，还允许开发者根据需求自由选择：利用 JAX 或 PyTorch 的即时执行模式进行高效调试，或切换至速度最快的后端以获得最高 350% 的性能提升。此外，Keras 具备强大的扩展能力，能无缝从本地笔记本电脑扩展至大规模 GPU 或 TPU 集群，是连接原型开发与生产部署的理想桥梁。",63927,"2026-04-04T15:24:37",[62,57,32]]