Skip to content

GAIA (Galaxy All-in-One Agent)

A unified natural-language interface for cross-Galaxy data analysis — an agentic AI front-end that lets researchers drive Galaxy tools, workflows, and histories through natural-language conversation, federating across Galaxy instances. The Goecks-lab counterpart to GalaxyMCP: GAIA is the user-facing agent, GalaxyMCP is the server-side protocol bridge.

  • Lead authors: Junhao Qiu, Jeremy Goecks (Cleveland Clinic — Goecks lab)
  • Galaxy tool / platform: TBD — verify from the talk slides (likely under goeckslab/ on GitHub)
  • Source: TBD — verify from the talk slides
  • Documentation / training: TBD — verify from the talk slides
  • Status at GBCC2025: oral talk on a natural-language interface for Galaxy

Talk at GBCC2025

  • Speakers: Junhao Qiu, Jeremy Goecks
  • Day / session: Day 2 (Tue Jun 24, 2025) — Oral Session 1, Grace Auditorium (chair: Sehyun Oh)
  • Talk title: "Galaxy All-in-One Agent: A Unified Natural Language Interface for Cross-Galaxy Data Analysis"
  • Slides: TBD
  • Video: GBCC2025 playlist
  • Abstract / schedule: GBCC2025 program

What it does

GAIA wraps Galaxy's tool registry, workflow engine, and history/dataset model behind an LLM-driven conversational layer. A user describes an analysis in English ("run variant calling on these PacBio reads, then annotate with VEP"); the agent picks tools, sets parameters, kicks off jobs, and surfaces results — federating across multiple Galaxy instances rather than locking the user to one server. The pitch is to lower the on-ramp for biologists who'd otherwise hit the Galaxy UI's tool-discovery cliff.

How it works

Core idea. GAIA is an agentic LLM front-end that translates natural-language requests into Galaxy API calls. The agent loop is: parse user intent, select tools/workflows from the Galaxy tool registry, fill parameters, invoke the BioBlend API, monitor histories, and report results — chained across turns so multi-step analyses build up in conversation.

Inputs / outputs. Input is a natural-language prompt (plus optional dataset references) and a target Galaxy instance (or set of federated instances). Output is one or more Galaxy histories with completed jobs, plus a conversational summary of what was run and where outputs live.

Key innovation. Cross-Galaxy federation in a single agent: the same chat session can run tools on usegalaxy.org, .eu, .org.au, or a private instance depending on data residency, tool availability, or load — the agent abstracts the "which server" decision away from the user.

Parameters / API surface worth knowing. - LLM backend — not specified in docs (talk-stage tool; repo / model choice TBD). - Galaxy connection layer — almost certainly BioBlend, the same Python client GalaxyMCP uses. - Tool / workflow selection — likely retrieval-augmented over the Galaxy ToolShed and IWC workflow registry. - History as memory — completed Galaxy histories provide the reproducibility / audit trail an LLM session would otherwise lack.

Canonical example. Not specified in public docs. The talk title frames the canonical interaction as a researcher asking GAIA to run an analysis spanning multiple Galaxy servers without manually choosing between them.

Where it fits in the corpus

  • AACR 2026: axis = agentic AI; this is the most-relevant non-AACR talk for the agentic-AI dossier — a working agent grounded in a real, large bioinformatics tool ecosystem rather than a benchmark
  • GalaxyMCP (entry) — sister talk, same authors, same session; GAIA is the agent, GalaxyMCP is the protocol substrate
  • EuroBioC 2025: no direct counterpart on the agentic-AI side
  • Nextflow Summit 2026: thematic overlap with the AI-for-pipelines storyline

Notes

GAIA + GalaxyMCP appear back-to-back in Oral Session 1 — together they're the AI/automation half of GBCC2025's "joint Galaxy + Bioconductor" rationale. Watch for whether the agent is a hosted service or a local-LLM-friendly client; the answer changes how reproducible an AI-driven Galaxy session can be.