OMIX Limma Analysis — Code Ocean Adapter

Fit limma linear or donor-blocked models to a declared continuous feature-by-sample matrix with aligned sample metadata.

Canonical OMIX module

Item	Location
Canonical module	OMIX Limma Analysis
Interface contract	schemas/interface.yml
Development contract	OMIX module contract
Version and source record	OMIX_MODULE_SOURCE.md

Canonical OMIX owns the scientific function, portable CLI, tests, and public input/output contract. Its complete R/ tree is exported byte-for-byte under code/functions/. This repository owns only Code Ocean file discovery, parameter translation, result placement, runtime selection, and adapter tests.

Main uses

Use this capsule for declared continuous matrices such as:

• donor-level means from corrected SCT/data expression;
• donor-level means from a declared corrected expression assay such as Harmony/data;
• GSVA or ssGSEA enrichment scores; or
• CIBERSORT relative fractions or absolute scores;
• normalized microarray, bulk-expression, proteomics, metabolomics, methylation M-value, chromatin-accessibility, CRISPR-score, qPCR delta-Ct, or copy-number log2-ratio matrices; or
• another documented continuous feature score with aligned biological replicates.

Do not use it for raw RNA counts. Raw-count pseudobulk must be analyzed with OMIX DEG Analysis so that library normalization and voom precision weights are applied. CIBERSORT relative fractions are supported through the declared fraction input type and transformation controls; do not pass other proportions as generic continuous scores without a justified transformation.

Inputs

Input	Requirement
Continuous matrix	CSV, TSV, TXT, or RDS feature-by-sample table. The default feature column is GeneName.
Sample metadata	CSV, TSV, TXT, or RDS table with a unique Sample column aligned to matrix sample columns.
Pseudobulk manifest	Optional Pseudobulk_Manifest.dcf from OMIX Seurat Pseudobulk. Required for that workflow handoff.

Each input can be selected explicitly. When a selector is blank, the adapter recursively examines attached files under /data and proceeds only when one compatible candidate is unambiguous. Multiple candidates are listed in the error so the user can select the intended file.

For an OMIX Seurat Pseudobulk result, keep these three files together:

• SCT_Mean_Log2_Expression.csv or Harmony_Mean_Expression.csv;
• Pseudobulk_Sample_Metadata.csv; and
• Pseudobulk_Manifest.dcf.

The manifest verifies the matrix semantics and downstream route. With Variance Model set to auto, Harmony means use ebayes and SCTransform means use ebayes_trend. A raw-count manifest stops with instructions to use OMIX DEG Analysis.

Run the analysis

1. Attach an upstream Pseudobulk result or provide the continuous matrix and metadata files directly.
2. Enter one or two contrast-variable columns. Supply one or more comma-separated Limma contrasts, such as B-A or 1-0 when the modeled groups are numeric, or leave the field blank only when the selected model has exactly two groups with at least two samples each. In that one unambiguous case, the capsule infers and records the sole comparison.
3. Add covariates only when scientifically justified. Add a donor variable only when a donor contributes repeated modeled profiles.
4. Select the scientific Input Data Type, or keep auto only when a compatible Pseudobulk manifest or backward-compatible Input Kind supplies the declaration. The program never guesses SCT, Harmony, GSVA, or another assay type from numeric values alone.
5. Keep Variance Model at auto to use the documented default for the declared data type, or select an explicit model after reviewing the variance diagnostic. other_continuous requires both an explicit variance model and a Measurement Scale description.
6. For CIBERSORT relative fractions, auto uses an arcsine-square-root transform. An explicit logit transform uses the Fraction Pseudocount at zero and one boundaries.
7. Run the capsule.

input_kind controls effect-size naming and normally remains auto. Log2 expression returns signed FC and logFC columns; enrichment and continuous scores return effect in the input units. See the canonical module guide for the variance-model recommendation for every supported data type.

Outputs

Output	Purpose
Limma_Analysis.csv	Contrast statistics and, by default, the aligned modeled matrix.
Sample_Metadata.csv	Metadata in modeled sample order.
run_summary.txt	Resolved inputs, variance model, design, contrasts, donor correlation, source, and runtime provenance.
Variance_Diagnostic.csv	Per-feature average signal and pre-moderation residual standard deviation.
variance_diagnostic.png	Visual check for a remaining mean-variance trend.

Environment and reproducibility

• Runtime profile: r-statistics
• Code Ocean environment: codeocean/omix-r-statistics:r4.4.3-bioconductor3.20-v1
• Public immutable image: ghcr.io/nidap-community/omix-r-statistics@sha256:1325722877fec5167d171aa766ddf7bbfd056e4999bf40fd8c1eabee495da667
• Source and release record: OMIX_MODULE_SOURCE.md

Record the adapter commit, selected parameters, input asset or checksums, and the Code Ocean run or release identity with scientific results.

Troubleshooting

Message	Resolution
No compatible matrix or metadata	Attach the required file or select it explicitly.
Multiple compatible candidates	Select the intended file in the App Panel or attach only one input bundle.
raw_integer_counts	Run the bundle with OMIX DEG Analysis instead.
No metadata IDs match matrix columns	Check sample naming and the Sample ID Column setting.
Donor has no repeated profiles	Leave Donor Variable Column blank for an ordinary linear model.
Blank contrast is ambiguous	Select a replicated model variable and provide the intended contrast. The error lists group replicate counts.
Contrast is not estimable	Confirm group labels, contrast spelling, replicate counts, and confounded covariates.

Numeric group labels such as 0 and 1 are supported directly. Enter the natural contrast 1-0; the adapter preserves that label in its result columns while the canonical module uses valid internal R design names. 1-0 is rejected when 1 and 0 are not actual modeled groups, so ordinary arithmetic cannot silently become a biological contrast.

For developers

Read AGENTS.md and OMIX_MODULE_SOURCE.md before editing. Scientific changes belong in canonical OMIX and must reach code/functions/ as a complete byte-identical managed export.

References and support

• Canonical module documentation
• limma User's Guide
• OMIX issue tracker