This plugin exposes Bioconductor clusterProfiler enrichment nodes as an
Autonomics File-to-File container family.
clusterprofiler_ora reads:
- a TSV with a unique gene-identifier column (default
gene); - a GMT pathway collection with at least one gene after each pathway name.
It calls clusterProfiler::enricher() and writes:
clusterprofiler_ora.tsvclusterprofiler_ora.json
The TSV retains the stable upstream result columns. The JSON report records parameters, input sizes, result count, and the installed clusterProfiler version. A zero-row enrichment result is valid.
Reads:
- a TSV rank table with unique
geneidentifiers and finite numericscorevalues (input order may be unsorted); - a GMT pathway collection.
It sorts scores decreasingly, calls clusterProfiler::GSEA(), and writes:
clusterprofiler_gsea.tsvclusterprofiler_gsea.json
The TSV retains the stable upstream result columns. The JSON report records parameters, input and eligible-set sizes, result count, and the installed clusterProfiler version. Pathways with no ranked genes are not eligible and a zero-row enrichment result is valid.
Both nodes use caller-supplied GMT collections and do not query online GO or KEGG services.
podman build --format docker \
-t localhost/autonomics/clusterprofiler:0.1.0 .
CLUSTERPROFILER_RUNNER=container ./test_clusterprofiler_ora.sh
CLUSTERPROFILER_RUNNER=container ./test_clusterprofiler_gsea.shFor a host R installation containing clusterProfiler, data.table, and
jsonlite, use:
CLUSTERPROFILER_RUNNER=local ./test_clusterprofiler_ora.sh
CLUSTERPROFILER_RUNNER=local ./test_clusterprofiler_gsea.shThe manifest references the digest-pinned image published at
ghcr.io/auto-nomics/autonomics/clusterprofiler.
Current local build:
- clusterProfiler 4.16.0
- data.table 1.18.6.1
- jsonlite 2.0.0
The nodes keep the hardened defaults: isolated networking and a read-only
image root filesystem. Explicit resource limits are 2 CPUs, 8 GiB memory,
512 PIDs, and 1 GiB shared memory. Only /work, /tmp, and /dev/shm are
writable at runtime.