Authors: Yuan Gao# (高源), Yizhou Luo#, Wenzhuo Li#, Yunquan Lan, Han Jiang, Yongcan Chen, Xiao Yi, Lihao Fu*, Min Yang*, Tong Si*
State Key Laboratory for Quantitative Synthetic Biology, Shenzhen Institute of Synthetic Biology, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences
定量合成生物学重点实验室，中国科学院深圳先进技术研究院合成生物学研究所

🌐 Live Demo / 在线体验: https://LabScriptAI.cn

[English]
Laboratory automation enhances experimental throughput and reproducibility, yet widespread adoption is constrained by the expertise required for robotic programming. Here, we developed LabscriptAI, a multi-agent framework enabling large language models (LLMs) to autonomously generate and validate executable Python scripts for protein engineering automation. LabscriptAI automated cell-free protein synthesis (CFPS) and characterization of 298 green fluorescent protein (GFP) variants designed by 53 teams from 5 countries in a student challenge. The top variant matched the performance of extensively optimized superfolder GFP (sfGFP) while exploring a distinct sequence space. Furthermore, LabscriptAI orchestrated distributed automation across biofoundry and fume hood-enclosed systems to engineer enzyme variants utilizing formaldehyde—a sustainable but hazardous substrate—identifying a double mutant with 7-fold enhanced catalytic efficiency. The platform implements rigorous safety measures, including biosecurity screening, physical containment, and human-in-the-loop oversight, to safeguard autonomous protein engineering. LabscriptAI democratizes laboratory automation by eliminating programming barriers while promoting responsible research practices.

[中文]
实验室自动化能够显著提高实验通量和可重复性，但机器人编程所需的专业知识限制了其广泛应用。为此，我们开发了 LabscriptAI，这是一个多智能体（Multi-agent）框架，能够赋能大语言模型（LLMs）自主生成并验证用于蛋白质工程自动化的可执行 Python 脚本。LabscriptAI 成功自动化了无细胞蛋白合成（CFPS）流程，并对来自 5 个国家 53 支学生队伍设计的 298 个绿色荧光蛋白（GFP）变体进行了表征。结果显示，表现最佳的变体在探索不同序列空间的同时，达到了经过广泛优化的超折叠 GFP（sfGFP）的性能水平。此外，LabscriptAI 还协调了跨越生物铸造厂和通风橱封闭系统的分布式自动化流程，用于工程化利用甲醛（一种可持续但有害的底物）的酶变体，最终鉴定出一个催化效率提高 7 倍的双突变体。该平台实施了严格的安全措施，包括生物安全筛查、物理遏制和“人在回路（Human-in-the-loop）”监管，以保障自主蛋白质工程的安全性。LabscriptAI 通过消除编程障碍普及了实验室自动化，同时促进了负责任的研究实践。

[English]
LabscriptAI employs a Multi-Agent Architecture orchestrated by LangGraph, separating high-level planning from low-level code implementation.

1. Task Planning Agent: Coordinates workflow processing. It translates natural language requests into structured Standard Operating Procedures (SOPs) and initiates human-in-the-loop dialogue for validation.
2. Code Graph Backend: An iterative code generation engine featuring:
   • Coder Agent: Generates initial Python scripts utilizing domain knowledge (RAG) and hardware profiles.
   • Reviewer Agent: Validates code using platform-specific simulators (Opentrons Simulator, PyLabRobot, pyFluent).
   • Precise Refactoring Engine (PRE): Instead of regenerating entire scripts upon error, PRE analyzes simulation logs to generate targeted SEARCH/REPLACE patches, ensuring convergence and stability.

[中文]
LabscriptAI 采用由 LangGraph 编排的多智能体架构，实现了高层任务规划与底层代码实现的解耦。

1. 任务规划智能体 (Task Planning Agent)：负责协调工作流处理。它将用户的自然语言需求转化为结构化的标准操作程序（SOP），并启动“人在回路”对话进行确认。
2. 代码图后端 (Code Graph Backend)：一个迭代式的代码生成引擎，包含：
   • 编码智能体 (Coder Agent)：利用领域知识（RAG）和硬件配置文件生成初始 Python 脚本。
   • 审查智能体 (Reviewer Agent)：调用平台特定的模拟器（Opentrons Simulator, PyLabRobot, pyFluent）验证代码。
   • 精准重构引擎 (Precise Refactoring Engine, PRE)：当验证出错时，PRE 不会重新生成整个脚本，而是分析模拟日志，生成针对性的 SEARCH/REPLACE 补丁，从而保证代码收敛和稳定性。

[English]
Traditional LLM-based code generation often struggles with "fixing one bug but introducing another" when regenerating full files. LabscriptAI's Precise Refactoring Engine (PRE) adopts a patching strategy. It locates the specific error block based on simulator feedback and applies a minimal edit, significantly reducing token usage and improving success rates.

[中文]
传统的基于 LLM 的代码生成在重新生成整个文件时，经常会遇到“修复一个 bug 但引入另一个 bug”的问题。LabscriptAI 的 精准重构引擎 (PRE) 采用补丁策略。它根据模拟器的反馈定位具体的错误代码块，并应用最小化的编辑，显著减少了 Token 消耗并提高了成功率。

[English]
LabscriptAI is hardware-agnostic, supporting major liquid-handling platforms through unified interfaces and adapter layers:

• Opentrons (OT-2 & Flex): Native support via Opentrons Python API v2.
• Tecan (Fluent & EVO): Supported via our custom-developed pyFluent library, compiling Python to .gwl worklists.
• Hamilton (Vantage & STAR): Integrated via the open-source PyLabRobot driver.

[中文]
LabscriptAI 具有硬件无关性，通过统一接口和适配层支持主流液体处理平台：

• Opentrons (OT-2 & Flex)：通过 Opentrons Python API v2 提供原生支持。
• Tecan (Fluent & EVO)：通过我们需要自研的 pyFluent 库支持，将 Python 代码编译为 .gwl 工作列表。
• Hamilton (Vantage & STAR)：集成开源驱动 PyLabRobot 进行支持。

Figure 2: Cross-platform validation of the iGEM plate reader calibration protocol on Opentrons, Tecan Fluent, and Hamilton Vantage.

[English]
In the CAPE (Critical Assessment of Protein Engineering) competition, LabscriptAI automated the screening of 298 student-designed GFP variants. It handled complex deck layouts, multi-step dilutions, and plate reader integration.

[中文]
在 CAPE (蛋白质工程关键评估) 竞赛中，LabscriptAI 自动化了 298 个学生设计的 GFP 变体的筛选工作。它成功处理了复杂的甲板布局、多步稀释以及酶标仪集成。

[English]

• Biosecurity Screening: Integration with the IBBIS Common Mechanism to screen DNA sequences for potential threats before synthesis.
• Physical Containment: Distributed automation strategies for handling hazardous reagents (e.g., Formaldehyde) in fume hood-enclosed robots.
• Human-in-the-loop: Mandatory human validation for critical SOPs and safety-sensitive operations.

[中文]

• 生物安全筛查：集成 IBBIS 通用机制，在合成前对 DNA 序列进行潜在威胁筛查。
• 物理遏制：针对危险试剂（如甲醛），采用在通风橱封闭机器人中进行的分布式自动化策略。
• 人在回路：关键 SOP 和安全敏感操作强制要求人类验证。

For detailed installation instructions, please refer to 📄 Installation Guide (docs/INSTALLATION.md). 详细安装指南请参阅 📄 安装文档 (docs/INSTALLATION.md)。

[English]
We use a dual-virtual-environment architecture to isolate Opentrons dependencies (which often conflict with modern AI libraries). [中文]
我们采用双虚拟环境架构来隔离 Opentrons 依赖（其经常与现代 AI 库发生冲突）。

# Windows PowerShell (One-click setup / 一键安装)
.\scripts\setup-uv.ps1

1. Start Backend / 启动后端:

   .\.venv\Scripts\Activate.ps1
   uv run python main.py

2. Start Frontend / 启动前端:

   cd labscriptAI-frontend
   npm run dev

3. Access / 访问: Open your browser at http://localhost:5173

• 🛠️ Installation & Troubleshooting / 安装与故障排除
• 📚 API Architecture / API 架构说明
• 🧪 pyFluent Library / pyFluent 库文档

If you use LabscriptAI in your research, please cite our work, thanks!: 如果您在研究中使用了 LabscriptAI，请引用我们的工作，感谢大佬！：

Yuan G, et al. "Autonomous liquid-handling robotics scripting through large language models enables accessible and safe protein engineering workflows." bioRxiv (2025): 2025-09.

Key Technologies used / 关键技术:

• LangGraph: Build resilient language agents as graphs.
• CodeAct: Executable Code Actions.
• PyLabRobot: Hardware-agnostic interface for liquid-handling robots.
• IBBIS: Common Mechanism for DNA Synthesis Screening.

Facing code problem please contact: gaoyuanbio@qq.com

State Key Laboratory for Quantitative Synthetic Biology, Shenzhen Institute of Synthetic Biology,CAS. 中国科学院深圳先进技术研究院合成生物学研究所，定量合成生物学重点实验室