Fragmented tools
Modern omics analyses combine many specialized packages, file formats, and parameter conventions.
Accepted by Patterns
BioMaster
Knowledge-guided bioinformatics workflow automation
A multi-agent system that turns natural language analysis goals into executable, checked, and recoverable bioinformatics workflows.
47/49benchmark tasks completed end to end
95.9%workflow execution success rate
102bioinformatics tools covered
4specialized agents
2retrieval knowledge bases
Why it matters
Bioinformatics workflows are hard to automate because the difficult part is not a single command.
Long dependencies
Failures in early preprocessing steps can silently break downstream analyses unless outputs are checked step by step.
Changing knowledge
New workflows and tools appear quickly, so BioMaster separates workflow knowledge from execution knowledge.
Overview
BioMaster coordinates planning, execution, debugging, and checking in one reproducible loop.
The system combines role-based LLM agents with Plan RAG and Execute RAG so that workflow design and tool-level command generation are supported by different kinds of domain knowledge.
Workflow design
Transforms a high-level biological objective into structured steps with expected inputs, outputs, and tools.
Script execution
Generates and runs tool-specific commands by retrieving usage details from Execute RAG.
Error recovery
Inspects execution failures, retrieves troubleshooting guidance, and retries corrected commands.
Output validation
Verifies that required files exist, are non-empty, and match the planned format before moving forward.