exploreDE is a Shiny app for exploring differential expression results from RNA-seq, proteomics and CRISPR screens, and for making publication-quality figures from them: ordination plots, heatmaps, volcano plots, box plots, GO and KEGG pathway views, ORA and GSEA results, and comparisons across contrasts.
It opens an AnnData .h5ad file in the exploreDE/1.0 layout, which build_explore_h5ad() writes from the results of any pipeline (DESeq2, edgeR, limma, prolfqua, SAINTexpress), and the results of the FGCZ SUSHI and prolfqua pipelines.
Documentation: https://fgcz.github.io/exploreDE/
exploreDE is submitted to Bioconductor. Until it is released there, install it from GitHub:
install.packages("BiocManager")
BiocManager::install("fgcz/exploreDE") # needs Bioconductor >= 3.23 (anndataR >= 1.2.3)
Once it is on Bioconductor:
BiocManager::install("exploreDE")
Optional: KEGG pathway views, GLM-PCA, and formatted group labels. Without these packages the app hides the feature or shows plain text.
BiocManager::install(c("pathview", "glmpca", "ggtext", "gridtext"))
library(exploreDE)
exploreDE() # landing page: upload a file or load the demo dataset
exploreDE("my_data.h5ad") # open a file directly
exploreDE(c("a.h5ad", "b.h5ad")) # compare the contrasts of several files
exploreDE(se) # open a SummarizedExperiment or AnnData object
The menu shows only the tabs your data supports: DE tabs need a DE table, pathway tabs need enrichment results.
vignette("exploreDE"): the app's tabs, and opening a dataset in Rvignette("bring-your-own-data", package = "exploreDE"): building a file from your own results: counts only, proteomics, RNA-seq with pathways, organisms without an annotation package, and unrecognised column names?build_explore_h5ad: the argument referencesystem.file("spec/exploreDE-h5ad-1.0.md", package = "exploreDE"): the .h5ad format, for writing files from Python or another toolsystem.file("skills/exploreDE-byod/SKILL.md", package = "exploreDE"): give this to an AI assistant to have it write the import script for your dataOnly counts, a sample table and the omics type are required; DE tables and pathway results are optional.
build_explore_h5ad(
counts, # features x samples matrix
out = "my_data.h5ad",
omics = "rnaseq", # or "proteomics", "crispr"
coldata = coldata, # one row per sample, rownames = colnames(counts)
rowdata = annotation, # optional: gene ids, symbols, GO terms, ...
de = list(Treat_vs_Ctrl = de_table) # optional: one table per contrast
)
You don't need to rename columns: a dictionary recognises the usual DESeq2, edgeR, limma, prolfqua and SAINTexpress names, and warns when it can't place one. The prolfquapp DEA's AnnData.h5ad (prolfquapp 2.10.x) opens as it is.
The Save / Load Session tab downloads your settings across every tab (widgets, the feature bucket, pathway selections and Multi DE datasets) as one .json file, and restores them on the same dataset later. It holds no data. Format: system.file("spec/exploreDE-session-2.md", package = "exploreDE").
On a server, ?data= opens a dataset from the link:
?data=<result directory>: a SUSHI result on FGCZ's gStore (LDAP-checked)?data=<percent-encoded https link to an .h5ad>: a file from an allowed source (by default APB Export jobs and public pStore; set EXPLOREDE_URL_SOURCES to change the list). Plain https:// only, no redirects, up to 1 GB.The LDAP check takes the user from the SHINYPROXY_USERNAME environment variable, read once per R process, so run the app behind ShinyProxy with one container per user. Without that variable (plain Shiny Server, for example) every gStore link is refused; the demo, uploads and https:// links still work.
exploreDE
AI coding agents helped write this package:
Assisted-by: AI coding agent trailer since 2 November 2026; earlier commits are not marked, so treat the whole code base as AI-assisted.h5ad builder and reader, and the session format; shinytest2 tests drive the app in a browser; R CMD check, BiocCheck, lintr and the browser tests run in CI on every pushProvenance and licence:
inst/examples/README.md
The maintainer is responsible for all code in this package, whatever helped write it, and for its reliability and maintenance.
See CITATION.cff.