Skip to content

GSVA Galaxy wrapper

Gene Set Variation Analysis (GSVA) wrapped as a Galaxy tool — a CoFest deliverable from GBCC2025 that takes one of the most-cited Bioconductor packages and packages it as a Galaxy tool, serving as the proof-of-concept for the auto-integration pipeline that automatically wraps Bioconductor packages for Galaxy.

  • Bioconductor source package: GSVA
  • Galaxy wrapper PR: see github.com/galaxyproject/tools-iuc (draft as of GBCC2025 CoFest)
  • Source (upstream): github.com/rcastelo/GSVA
  • Documentation / training: GSVA vignette; Galaxy tool docs TBD on merge
  • CoFest authors: Fabio Cumbo, Jayadev Joshi, Bryan Raubenolt, Daniel Blankenberg (lead); on-site team Maria Doyle, Gretta Yagudayeva, Charlotte Soneson, Marcel Ramos Pérez; remote — Robert Castelo (GSVA maintainer, UPF)
  • Status at GBCC2025: CoFest deliverable / proof-of-concept for the auto-wrap pipeline

CoFest output at GBCC2025

  • Format: GBCC2025 Community Fest (CoFest) deliverable, not a session talk
  • Day / session: GBCC2025 CoFest, Jun 27-28, 2025 (post-conference, CSHL)
  • Title: "GSVA Galaxy wrapper"
  • Slides / output: PR on galaxyproject/tools-iuc (link to be added on merge)
  • Video: not recorded
  • Abstract / schedule: GBCC2025 program

What it does

GSVA computes per-sample gene-set enrichment scores from an expression matrix — turning a genes-by-samples matrix into a gene-sets-by-samples matrix that downstream tools can treat as features. The Galaxy wrapper exposes this as a tool with the same arguments (method = gsva / ssgsea / plage / zscore, kernel choice, min/max set size) so a Galaxy user can run GSVA on a workflow output without writing R. The deeper goal is the auto-wrap pipeline behind it: GSVA is the visible test case demonstrating that the Bioconductor-to-Galaxy pipeline works end-to-end on a non-trivial package.

How it works

Core idea. The Galaxy tool is a thin XML wrapper around the upstream GSVA::gsva() Bioconductor entry point. The wrapper declares the tool's parameters, points the Galaxy job runner at a Conda environment with R + Bioc + GSVA installed, runs an R driver script that loads the user's expression matrix and gene sets, and surfaces the resulting matrix as a Galaxy dataset.

Inputs / outputs. Input: an expression matrix (genes × samples, microarray or RNA-seq) and a gene-set collection (e.g. MSigDB GMT). Output: a gene-sets × samples score matrix that downstream Galaxy tools (clustering, differential analysis, survival modelling) treat as features.

Key innovation. GSVA performs a non-parametric coordinate-system change from genes to gene sets — preserving per-sample resolution while collapsing thousands of genes into hundreds of pathway scores, unsupervised and reference-free. As a Galaxy wrapper, the innovation is the auto-wrap pipeline: the initial wrapper was drafted by an LLM (Gemini) from the GSVA package's R-side metadata, then refined by humans — the CoFest deliverable proves the LLM-assisted bioc-to-galaxy path works on a real, widely-cited package.

Parameters / API surface worth knowing. - method — one of gsva, ssgsea, plage, zscore (algorithm choice). - kernel — controls the ranking approach for the non-parametric enrichment calculation. - min.sz / max.sz — gene-set size filters (drop sets outside the range). - Conda dependency — R + bioconductor-gsva pinned in the wrapper's <requirements> block.

Canonical example. From the GSVA vignette: load an ExpressionSet (e.g. a TCGA RNA-seq subset), pass MSigDB Hallmark gene sets, call gsva(expr, gene_sets, method = "gsva"), get a Hallmark-by-sample matrix. The Galaxy wrapper exposes exactly this call as a form: upload matrix, upload GMT, pick method, click run, download the score matrix.

Where it fits in the corpus

  • AACR 2026: axis = bioinfo-tools (gene-set enrichment is a default feature-engineering step for any cancer-omics analysis)
  • bioc-to-galaxy (entry) — GSVA is the proof-of-concept test case; success here unblocks a long tail of Bioc packages
  • atena (entry) — also Castelo's package; the GSVA wrapper closes the loop with the package maintainer remotely
  • Workshop Platform (cross-vault) — wrapped tools are easier to slot into hands-on tutorial environments

Notes

GSVA being one of the most-cited Bioconductor packages makes it a high-visibility test case — a successful wrap is broadly useful and gives the auto-pipeline an immediate poster child. Castelo's remote participation is the maintainer-loop closure: an auto-wrap that the upstream maintainer endorses is qualitatively different from one done over the maintainer's head.