# Gene-expression signatures

Source: https://onco.cc/terms/gene-expression-signatures/  
OnCo record `gene-expression-signatures` (Term). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

An expression signature combines measured gene messages using a defined scoring method.

## Summary

A signature combines selected RNA measurements by a defined algorithm; preprocessing, weights, reference data and software version are part of the definition; a gene list alone is not a test.

## Fields

- Kind: Term
- Last checked: 2026-11-13

## Notes

- A signature combines selected RNA measurements by a defined algorithm; preprocessing, weights, reference data and software version are part of the definition; a gene list alone is not a test.
- Pattern or module score, subtype classifier and outcome predictor answer different questions; performance belongs to a population, endpoint and horizon.
- The original PAM50 work used a defined 50-gene method for intrinsic subtypes and risk models and does not validate every later similar score.
- Seurat's AddModuleScore subtracts matched control-gene expression from the selected-gene mean; GSEA tests accumulation toward one end of a ranked list; outputs cannot share a cutoff.
- A clinical predictor needs a locked procedure; in the Venet et al. (2011) breast-cancer analysis, more than 90% of random signatures over 100 genes were significant outcome predictors, so association alone does not show the named biology matters.
- Analytical repeatability, clinical validity and clinical utility are separate questions.
- A bulk inflammatory signature can rise from more immune cells, changed state or both; a stress score does not establish protein activation, metabolic rate or dependence.
- Fictional teaching exercise: Replacing normalisation while retaining the old high cutoff is not validated; the old cutoff cannot supply missing validation.
- An expression score requires compatible RNA input, locked preprocessing, genes, weights, reference, endpoint/horizon and software version. Existing data can be rescored without new tissue; acquiring RNA consumes material. Report categories, ranks, probabilities or unitless scores on their exact scale. Missing genes, batch, mixture and population shift undermine cutoffs; analytical, clinical and utility evidence remain separate.
- An expression-program score has meaning within its algorithm and background. For a score formed by subtracting matched control-gene expression, zero need not mean that no program genes were detected. The 90th percentile places a result relative to a specified reference; it does not mean that 90 percent of cells express the program or that a pathway is 90 percent active. Changing the genes, preprocessing, controls or reference can change the scale, so retain them with the method and version.

## Sources

- Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles.: https://pubmed.ncbi.nlm.nih.gov/16199517/
- Supervised risk predictor of breast cancer based on intrinsic subtypes.: https://pubmed.ncbi.nlm.nih.gov/19204204/
- Most random gene expression signatures are significantly associated with breast cancer outcome.: https://pubmed.ncbi.nlm.nih.gov/22028643/
- Criteria for the use of omics-based predictors in clinical trials: explanation and elaboration.: https://pubmed.ncbi.nlm.nih.gov/24228635/
- Seurat AddModuleScore definition: https://satijalab.org/seurat/reference/addmodulescore

## Connected records

- terms: [Bulk RNA sequencing (RNA-seq) and the full transcriptome](https://onco.cc/terms/bulk-rna-seq/)

---
JSON: https://onco.cc/api/v1/entities/gene-expression-signatures.json