Limma Analysis

limma-analysis performs differential analysis of a pre-normalized, continuous feature-by-sample matrix with the canonical OMIX Limma module. It is suitable for normalized bulk expression, log2 microarray expression, pathway scores, cell-fraction estimates, methylation M-values, copy-number log2 ratios, and other declared continuous matrices. It is research software, not clinical decision support.

The standalone adapter validates the input boundary, invokes the canonical Limma CLI unchanged, and produces a portable result bundle. Canonical source commit and file hashes are recorded in OMIX_MODULE_SOURCE.md.

Scope

This release candidate provides only the primary continuous-matrix analysis. It does not provide the canonical cell-type collection entrypoint or the cell-type collection to L2P Multi workflow; those need a separate, versioned extension.

Inputs

Provide two aligned tables:

Input	Requirement
Matrix	CSV, TSV, TXT, or RDS feature-by-sample matrix. It must contain a feature identifier column (default GeneName) followed only by finite numeric sample columns.
Metadata	CSV, TSV, TXT, or RDS sample metadata. It must contain a sample ID column (default Sample) matching the matrix sample columns exactly, without duplicates.
Input type	An explicit --input_data_type selected from the adapter contract.

Raw integer counts are rejected. This adapter does not determine whether values look like counts, normalize them, or transform them: the declared data type is the caller’s scientific contract. Use a count-aware workflow before calling this adapter when starting from raw counts.

Validated image

The approved v1.0.0 Linux/AMD64 image was built from source commit 8fe2cd7eea6f22cb14eabe78ff6b4f79d2ea5ef0. Use its immutable digest for reproducible execution:

ghcr.io/nidap-community/limma-analysis@sha256:83d73524eac2d346b329e20f4de5e7bb330c1be96a31a8afe331883280a193a2

v1.0.0 and latest are readable tags that resolved to this digest at publication. The versioned GitHub Release is published as v1.0.0.

Build

The container uses an immutable AMD64 base image digest. Build a local image:

container/build_docker.sh limma-analysis:local

Run

Place matrix.tsv and metadata.tsv in an input directory. The provided helper mounts that directory read-only at /data and mounts a new output directory at /output.

bash container/run_docker.sh limma-analysis:local ./input ./results \
  --matrix /data/matrix.tsv \
  --metadata /data/metadata.tsv \
  --gene_names_column GeneName \
  --sample_names_column Sample \
  --contrast_variable_columns Group \
  --contrasts B-A \
  --input_data_type normalized_bulk_expression \
  --output_dir /output

--contrast_variable_columns and --contrasts define the comparison. The canonical command also supports covariates, paired designs, batch columns, and other documented Limma options; view them with:

docker run --rm --platform linux/amd64 limma-analysis:local --help

Results

The canonical result files remain at the output root for compatibility. Stable copies make downstream use predictable:

results/
├── tables/
│   ├── Limma_Analysis.csv             # inferential differential-analysis results
│   ├── Sample_Metadata.csv            # modeled sample metadata
│   ├── Variance_Diagnostic.csv
│   └── Batch_Corrected_Expression.csv # optional downstream matrix, not inference
├── figures/
│   └── variance_diagnostic.png
├── logs/
│   ├── run_summary.txt
│   └── sessionInfo.txt
├── RUN_METADATA.md
└── run_manifest.json

run_manifest.json is written only after the canonical CLI exits successfully. When batch-corrected values are requested, they are deliberately kept separate from Limma_Analysis.csv, because batch correction is a downstream matrix and not a replacement for model-based inference.

Synthetic fixture example

The small redistributable fixture in test_data/ has three A and three B samples. For the declared B-A contrast, its first three deliberately shifted features are returned as follows:

Feature	B-A log fold change	Adjusted p-value
Gene1	1.9	3.37e-17
Gene2	1.8	5.83e-17
Gene3	1.7	1.44e-16

This is a synthetic validation fixture, not a biological finding. Its expected output and visual inspection notes are recorded in docs/README_EXAMPLE_PROVENANCE.md.

Validate locally

container/smoke_test.sh limma-analysis:local
container/parity_test.sh limma-analysis:local
bash tests/test_adapter_contract.sh limma-analysis:local

The smoke test checks the complete result bundle. The parity test compares canonical and adapter outputs on the same fixture. The contract test confirms that undeclared types, raw-count manifests, and misaligned samples fail without creating a success manifest.

Troubleshooting

• --input_data_type is required or auto is rejected: state the known continuous assay type explicitly; the adapter will not guess it.
• Raw-count manifest is rejected: normalize counts in a count-aware workflow first, then provide the resulting declared continuous matrix.
• Sample identifiers do not align: make the metadata sample IDs exactly match the matrix sample columns, including spelling and uniqueness.
• Output directory is not empty: select a new output directory. This avoids confusing an earlier success manifest with the current run.

HPC use

For Apptainer or Singularity, first build or obtain a local SIF from an approved image, then use the same input/output layout:

bash container/run_apptainer.sh limma-analysis.sif ./input ./results primary \
  --matrix /data/matrix.tsv --metadata /data/metadata.tsv \
  --gene_names_column GeneName --sample_names_column Sample \
  --contrast_variable_columns Group --contrasts B-A \
  --input_data_type normalized_bulk_expression --output_dir /output

See docs/container_contract.md for the portable runtime boundary and docs/architecture.md for the canonical/adapter separation.