Skip to content

signifinder

Compute and explore public cancer transcriptional signatures — a Bioconductor package containing 70+ curated tumor signatures from the literature and applying them as single-sample / single-cell / single-spot scores across bulk RNA-seq, microarray, single-cell, and spatial transcriptomics inputs.

Talk at GBCC2025

  • Speakers: Stefania Pirrotta, Laura Masatti, et al. (Calura Lab, University of Padua)
  • Day / session: Day 2 (Tue Jun 24, 2025) — Oral Session 2B, Grace Auditorium (chair: Vincent Carey)
  • Talk title: "Public gene expression cancer signatures across bulk, single-cell and spatial transcriptomics data with signifinder Bioconductor package"
  • Slides (Zenodo DOI): TBD — Zenodo deposits typically 2–4 weeks post-conference
  • Video: GBCC2025 playlist
  • Abstract / schedule: GBCC2025 program

What it does

signifinder is a curation play: 70+ published cancer signatures (hypoxia, EMT, immune-infiltrate, stemness, and dozens of indication-specific scores) packaged with one consistent scoring API that runs across bulk RNA-seq, microarray, single-cell, and spatial inputs. Instead of every analyst re-implementing each signature from a paper's supplement, signifinder has them in one place with the same input contract — so a user can compute and compare a panel of signatures on the same object.

How it works

Core idea. Each bundled signature is re-implemented in its original published form — some as weighted sums of gene-level coefficients (e.g. Pyroptosis_Ye uses per-gene weights in the −0.187 to 0.130 range), others as enrichment scores via ssGSEA or GSVA, depending on what the source paper specified. The scoring is per-sample (or per cell / per spot), so the same code path runs across modalities.

Inputs / outputs. Input is normalized RNA-seq counts or microarray data wrapped in a SummarizedExperiment, SingleCellExperiment, or SpatialExperiment; gene IDs can be symbols, Entrez, or Ensembl. Output is the same object with one column per signature appended to colData.

Key innovation. A single API spanning bulk, single-cell, and spatial — earlier signature collections targeted bulk only — plus an evaluation layer that reports the fraction of signature genes actually present in the dataset (e.g. "is using 100% of signature genes").

Parameters worth knowing. - inputType — "rnaseq" or "microarray", switches normalization assumptions. - nametype — gene-ID convention used in the assay. - whichSign / tissue / tumor / topic — filters in multipleSign() for computing a batch of signatures at once. - whichAssay — which assay slot to score when the object holds multiple.

Canonical example. The vignette loads ovse, an ovarian-cancer bulk RNA-seq SummarizedExperiment, and calls ferroptosisSign(dataset = ovse, inputType = "rnaseq"). The returned object carries the ferroptosis score in colData, and the same object is then chained into further signature calls (e.g. EMTSign, ImmunoScore_Hao) so a panel of scores accumulates on one object. availableSignatures() and evaluationSignPlot() support browsing and QC.

Where it fits in the corpus

  • AACR 2026: axes single-cell-spatial-omics + clinical-trials — cancer signatures are clinically-actionable readouts and the cross-modality coverage matches the AACR spatial / single-cell dossiers
  • mitology (entry) — same speaker (Pirrotta), Day 4; sister package on the same "transcriptome-interpretation" track
  • ISMB 2026: scaffold
  • Nextflow Summit 2026: not mentioned

Notes

Pirrotta presents twice at GBCC2025 — signifinder on Day 2 and mitology on Day 4 — which signals the Calura lab is consolidating a "transcriptome interpretation" suite (signature scoring + pathway / organelle-specific scoring) rather than shipping one-off packages. The cross-modality angle (bulk + sc + spatial) is the key value-add over older signature collections that targeted bulk only.