CellAgent: An LLM-driven Multi-Agent Framework for Automated Single-cell Data Analysis

单细胞RNA测序（scRNA-seq）数据分析对于生物研究至关重要，因为它能够精确地表征细胞异质性。然而，手动操作各种工具以实现预期结果可能会使研究人员的工作量很大。为了解决这个问题，我们介绍了CellAgent（http://cell.agent4science.cn/），这是一个LLM驱动的多代理框架，专门设计用于自动处理和执行scRNA-seq数据分析任务，提供高质量的结果，无需人工干预。首先，为了使通用LLM适应生物领域，CellAgent构建了LLM驱动的生物专家角色——规划者、执行者和评估者——每个角色都有特定的职责。然后，CellAgent引入了一个分层决策机制来协调这些生物专家，有效地推动复杂数据分析任务的规划和逐步执行。此外，我们提出了一个自我迭代优化机制，使CellAgent能够自主评估和优化解决方案，从而保证输出质量。我们在一个包含数十种组织和数百种不同细胞类型的全面基准数据集上评估了CellAgent。评估结果一致表明，CellAgent有效地识别了单细胞分析任务的最合适的工具和超参数，实现了最佳性能。这个自动化框架大大减轻了科学数据分析的工作量，使我们进入了“科学代理人”时代。

CellAgent: An LLM-Driven Multi-Agent Framework for Automated scRNA-Seq Data Analysis

CellAgent is a multi-agent framework that automates scRNA-Seq data analysis tasks using a hierarchical decision-making mechanism and a self-iterative optimization mechanism, resulting in high-quality results with no human intervention.

CellAgent introduces LLM-driven biological expert roles - planner, executor, and evaluator - and uses a hierarchical decision-making mechanism to coordinate them. It also has a self-iterative optimization mechanism that guarantees output quality. The framework was evaluated on a comprehensive benchmark dataset and consistently achieved optimal performance. The automated framework reduces workload for data analyses.

Related studies in this field include 'Single-cell RNA sequencing: advances and future challenges' and 'A comparison of single-cell trajectory inference methods'.