BioMaster: Multi-agent System for Automated Bioinformatics Analysis Workflow

“BioMaster [38] employs a similar multi-agent design, where specialized agents collaborate to plan, code, and debug, enabling seamless execution of complex tasks such as Hi-C data processing.”

“While existing multi-agent systems primarily rely on large language models 24,32,33 , we leveraged a smaller, more efficient language model, Phi-3 34 .”

“By leveraging LLMs, it is possible to create intuitive applications that allow users to query and analyze scRNA-seq data using natural language[17, 18, 19, 20, 21], thereby democratizing access to advanced analytical tools.”

“The advent of large language models (LLMs) is transforming bioinformatics by enabling natural language–driven workflows that reduce technical barriers and simplify complex analyses [1].”

“Recent work has shown that LLM agents can interface with these tools to generate candidate protein designs from natural language specifications [20–22], suggesting a path toward AI-driven molecular design pipelines.”

“Large language model (LLM) agents have recently found broad applications across scientific domains, including biology [34], [35], [36], [37], [38], chemistry [39], [40], [41], materials science [42], [43], [44], and physics [45], [46].”

“Other approaches like BioMaster [14] and GeneAgent [15] have since emerged to explore the application of agentic systems to tackle the challenge.”

“A dedicated Debug Agent monitors the pipeline for errors or suboptimal results (such as a file format issue or poor sequence coverage) and can intervene to adjust parameters or retry steps when needed [32].”

“Recent agent frameworks such as AutoBA [59] and BioMaster [45] rely on small-scale, human-curated datasets, which require domain experts to construct and incur high computational costs for large-scale evaluation.”

“When the searches were updated, an increase in the number of Artificial Intelligence (AI) and chat-based tools was apparent, but most were intended for exploratory analysis rather than workflow creation (e.g. RiboChat, GeneGPT, BioMaster [ 42– 44].”

“In contrast to traditional workflow automation, MAS-based frameworks can explicitly model domain knowledge, reason over task dependencies, and dynamically construct executable processes based on user objectives and data characteristics 14 .”

“BioMaster (Su et al., 2025) enhances long-horizon workflow execution by integrating RAG and optimizing agent coordination for extended pipelines.”

“Other recent frameworks [29, 16] investigate multi-agent architectures for automating bioinformatics workflows, tool orchestration, error recovery, and computational analysis pipelines.”

“Existing evaluations typically assess step-level capabilities on curated tasks, such as BixBench [15], or whether an agent can construct an appropriate workflow, as in BioMaster [16].”

“Multi-agent architectures such as BioMaster [27], BioAgents [18] and Biomni [10] enable inter-agent coordination, planning and task execution to nominate hypotheses for validation.”

“BioMedAgent obtains 10.3% and 3.1% under the two correctness metrics, BixBench obtains 0.8% under each, and BioMaster and Biomni obtain no correct answers.”

“BioMaster, GeneAgent, and ReflecTool further show that biomedical agents need planning, retrieval, tool use, and self-verification (Su et al., 2025; GeneAgent, 2025; Liao et al., 2024).”

“BioMANIA [12] focused on an inquiry-based assistant for processing omics data, while BioAgents [13] and BioMaster [4] leveraged multi-agent architectures to decompose complex genomic tasks into manageable code generation steps and CellAgent[7] uses a plan,execute and evaluate style multi-agent…”

“Um dieser Konstellation zu begeg-nen, haben verschiedene Gruppen [44, 49, 57, 62, 64]”

“BixBench [6] grades closed models on more than 50 Dockerised computational-biology scenarios at ~17 % open-answer accuracy; BioMaster [7] wraps a Plan/Task/Debug/Check loop around retrieval-augmented generation (RAG) for RNA-seq, ChIP-seq, scRNA-seq, and Hi-C analyses; BioAgents [8] fine-tunes…”

“Broader biological agent platforms include Virtual Lab [20], Biomni [21], BioLab [22], OriGene [23], and Medea [24], alongside systems for bioinformatics workflow automation [25, 26], self-verified gene-set analysis [27], and knowledge-graph construction [28].”

“In computational biology, early LLM-based systems have demonstrated tool use and workflow assistance in domains such as genome analysis and transcriptomics Zhou et al. (2024); Huang et al. (2025); Su et al. (2025).”

“BioMaster integrates retrieval-augmented generation and a dedicated check agent to enable robust execution of complex workflows, including ChIP-seq and Hi-C analyses, through dynamic knowledge retrieval and stringent input-output validation [16].”

“Systems such as AutoBA [20], BioMaster [21], and Biomni [14] en-386 code knowledge of which tools to run and in what order for specific scientific domains.”

“For instance, systems like BioAgents [9] and the recently proposed BioMaster [17] employ multi-agent architectures to assist researchers in executing multi-step bioinformatics pipelines, ranging from RNA-seq analysis to genome assembly, often through natural language commands.”

“…Hyb SA GG (Nouri et al., 2025) Agentic AI SA Hyb SA GG (Martinek et al., 2025) Agentomics-ML SA Hyb A GG (Huang et al., 2025) Biomni SA Hyb SA GG (Su et al., 2025) BioMaster H Seq SA GG (Raghavan et al., 2025) CellAgentChat D Hyb A GG (Mehandru et al., 2025) BioAgents HY Hyb SA ONC (Liu et al.,…”