Skip to content

gINTomics

Multi-omics data integration and visualization — a Bioconductor package that identifies associations between a target gene's expression and its regulatory factors (Copy Number Variations, methylation), bundling the integration models with a Shiny app for interactive exploration of the results.

Talk at GBCC2025

  • Speakers: Angelo Velle, Francesco Patanè, Chiara Romualdi (University of Padua)
  • Day / session: Day 2 (Tue Jun 24, 2025) — Oral Session 2A, Bush Hall (chair: Jenny Drnevich)
  • Talk title: "gINTomics, a powerful Bioconductor package for multiomics data integration and visualization"
  • Slides (Zenodo DOI): TBD — Zenodo deposits typically 2–4 weeks post-conference
  • Video: GBCC2025 playlist
  • Abstract / schedule: GBCC2025 program

What it does

gINTomics ties expression data to its regulatory layer — CNV and methylation — and tests for associations gene by gene. The output is a per-target model surfaced through a Shiny-based app, so the integration step and the visualization step ship together rather than expecting the user to wire one to the other. The Romualdi lab's positioning is "powerful + visual" rather than "novel statistics."

How it works

Core idea. For each target gene, gINTomics fits a regression of expression against its candidate regulators — CNV at the gene locus, methylation, miRNAs, and TFs — choosing a negative-binomial model for RNA-seq counts or a linear model for microarray / continuous data. When the regulator set is large, a random-forest pre-selection prunes to the most important covariates before the regression is fit.

Inputs / outputs. Inputs are a MultiAssayExperiment carrying any combination of gene expression (RNA-seq or microarray), miRNA expression, methylation, gene CNV, and miRNA CNV. Regulator dictionaries are pulled automatically: TF–target and miRNA–target relations via OmnipathR, TF–miRNA via TransmiR. Output is a per-target list of integration models, which is then fed directly into the Shiny visualization layer.

Key innovation. Pairing the integration step with a packaged Shiny app surfaced by run_shiny() — so the same object that holds the regression results is also the input the GUI consumes, rather than requiring the analyst to bridge between modeling output and a separate visualization stack.

Parameters worth knowing. - Data-type switch — model family flips between negative binomial (counts) and linear (microarray); chosen from input type. - Regulator-dictionary source — OmnipathR for TF / miRNA targets, TransmiR for TF→miRNA edges; affects which covariates enter each gene's model. - Random-forest pre-selection — engaged when the regulator count is high; trims covariates before regression. Exact threshold not specified in vignette; needs talk slides.

Canonical example. The vignette loads mmultiassay_ov, a 20-patient TCGA ovarian-cancer MultiAssayExperiment carrying expression, miRNA, methylation, and gene/miRNA CNV. Users extract the individual matrices, rebuild a clean MultiAssayExperiment, and run run_multiomics() to fit the integration models; the resulting object is passed directly to run_shiny() to launch the interactive viewer.

Where it fits in the corpus

  • AACR 2026: axis bioinfo / AI methods — gINTomics is one of the multi-omics-integration tools relevant to the cancer-genomics dossiers
  • MOSClip (entry) — same lab family; both are multi-omics integration packages, but MOSClip is survival-focused while gINTomics is regulator-association-focused with explicit visualization
  • ISMB 2026: scaffold
  • Nextflow Summit 2026: not mentioned

Notes

Mature package (Bioconductor 3.23, v1.8.0). The differentiator versus MOSClip and the broader multi-omics ecosystem is the embedded Shiny app — Romualdi's group is consolidating an "integration + interactive exploration" workflow rather than shipping a bare modeling library.