AI4SCIENCE
AI for Science：课题组的机器学习模型、分析流程与数据集，以及我们使用 AI 做科研的九条组内经验。 AI for Science: machine-learning models, pipelines, and datasets from the lab, plus nine practice notes on how we work with AI.
资源Resources
Protocols、工具与数据库资源汇总。 Protocols, tools, and database resources.
组内 AI 使用经验Working with AI — Lab Notes
课题组在实验设计、数据分析、写作与基金申请中使用 Claude Code、Claude Science、Codex 与 Kimi 的实践经验，欢迎每位组员补充。 Practice notes on using Claude Code, Claude Science, Codex, and Kimi for experimental design, data analysis, writing, and grants. Every member is welcome to add one.
SAIR 视频总结SAIR Video Summaries
陶哲轩等创立的「科学与 AI 研究基金会」(SAIR) 全部讲座、圆桌与访谈的中文深度总结，共 57 个视频。 Chinese deep-dive summaries of all 57 talks, panels, and interviews from SAIR (Foundation for Science and AI Research, co-founded by Terence Tao).
Claude Science
Anthropic 面向科研的 AI 工作台：60+ 生物学数据库、单细胞与基因组分析、蛋白结构预测。 Anthropic's AI workbench for research: 60+ biology databases, single-cell and genomics analysis, protein structure prediction.
实验 ProtocolsLab Protocols
细胞培养、甲基化检测、表观遗传编辑等实验流程。 Cell culture, methylation assays, epigenome editing, and more.
生物信息学工具Bioinformatics Tools
分析脚本、pipeline、可复用代码库。 Analysis scripts, pipelines, and reusable code repositories.
常用数据库Public Databases
常用公共数据库入口： Frequently used public databases:
推荐阅读Recommended Reading
入门教程与领域综述文章合集。 Introductory tutorials and review articles in our field.
AI 科研工具AI Research Tools
Claude Science
Anthropic 于 2026 年 6 月发布的科研工作台（AI workbench），把文献、数据库、计算流程与作图写作整合到同一个环境中。对生物学研究而言，它的价值不在于「又一个聊天模型」，而在于把分散在 UniProt、PDB、Ensembl、GEO 等资源与本地 pipeline 之间的往返工作串成一条可复现的链路。 Claude Science is Anthropic's AI workbench for researchers, released in June 2026. It brings literature, databases, compute pipelines, and figure/manuscript writing into one environment. For biology, its value is less "another chat model" than stitching the round trips between UniProt, PDB, Ensembl, GEO and local pipelines into one reproducible chain.
跨库检索与综合Cross-database Retrieval
内置 60 余个科学数据库与连接器，涵盖 UniProt、PDB、Ensembl、Reactome、ClinVar、ChEMBL、GEO 等。用自然语言提出一个问题，由专门的子代理分别查询各库并汇总结果，再由 reviewer 代理复核。文献侧接入 PubMed、bioRxiv 与 ClinicalTrials.gov。 More than 60 scientific databases and connectors are built in, including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO. Ask a question in plain language and specialist sub-agents query each source and synthesise the answer, with a reviewer agent checking the result. Literature access covers PubMed, bioRxiv, and ClinicalTrials.gov.
面向生物学的预配置Domain Presets for Biology
针对基因组学、单细胞、蛋白质组学与化学信息学做了预配置，并附带成套 Agent Skills，例如单细胞 RNA-seq 的质控技能（过滤阈值、UMAP 可视化）、实验与临床方案生成、通路富集作图与跨研究比较。常见用途包括 scRNA-seq 分析、CRISPR 靶点设计、蛋白折叠与化学信息学筛选。 Pre-configured for genomics, single-cell, proteomics, and cheminformatics, with bundled Agent Skills such as scRNA-seq quality control (filtering thresholds, UMAP plots), experimental and clinical protocol drafting, and pathway-enrichment charting across studies. Typical uses include scRNA-seq analysis, CRISPR guide design, protein folding, and cheminformatics screening.
专业模型与集群调度Specialist Models and HPC
可调用 NVIDIA BioNeMo 系列模型（Evo 2、Boltz-2、OpenFold3），也能接入课题组自有 pipeline 与私有工具。计算方面支持向 HPC 集群提交作业、SSH 远程执行或使用 Modal 的按需算力，从单卡扩展到数百卡都无需离开工作台。 It can call NVIDIA BioNeMo models (Evo 2, Boltz-2, OpenFold3) and also plug into a lab's own pipelines and proprietary tools. For compute it submits jobs to HPC clusters, runs over SSH, or uses Modal's on-demand capacity, scaling from one GPU to hundreds without leaving the workbench.
图表、代码与出处同时保留Figures with Code and Provenance
生成图表与稿件时会一并保留产生它们的代码、运行环境与完整对话记录，引用带可核查的出处。界面可原生渲染 3D 蛋白结构、基因组浏览器轨迹与化学结构式，便于在组会上直接讨论中间结果。 Figures and manuscripts are produced together with the code, environment, and full message history behind them, and citations are traceable. The interface natively renders 3D protein structures, genome-browser tracks, and chemical structures, so intermediate results can be discussed directly in lab meetings.
获取方式与延伸阅读Access and Further Reading
目前处于 beta 阶段，面向 Claude Pro、Max、Team 与 Enterprise 用户，支持 macOS 与 Linux；学术机构与非营利组织可申请折扣版 Team 计划。此前发布的 Claude for Life Sciences（2025 年 10 月）提供 10x Genomics Cloud、Benchling 等平台连接器。Anthropic 的 AI for Science 计划则面向有研究机构背景、从事高影响力课题（尤其是生物与生命科学方向）的研究者提供支持。 Currently in beta for Claude Pro, Max, Team, and Enterprise users on macOS and Linux; academic institutions and nonprofits can apply for a discounted Team plan. The earlier Claude for Life Sciences release (October 2025) added connectors for platforms such as 10x Genomics Cloud and Benchling. Anthropic's AI for Science program supports researchers at research institutions working on high-impact projects, with a focus on biology and the life sciences.