Concept
Expression-Based Prognostic Signatures
5 studiesEvidence last moved Sep 20, 2026
An expression-based prognostic signature is a set of genes whose measured activity is used to predict how a patient's disease will progress. Almost all of them are discovered by mining public expression repositories, which shapes both what they can find and what they can show.
These studies are numerous and easy to produce, and they all end with a survival curve that separates. The question a reader has to ask is whether the separation was discovered or constructed, because the same pipeline — pick genes correlated with outcome, then test that they correlate with outcome — produces a good-looking curve from data with no signal in it.
Studies
5
Findings
4
5 supporting · 0 challenging · 0 qualifying citations
Open tensions
1
Latest change
Concept page published
Expression-Based Prognostic Signatures
Currently
What we know
- Both markers track immune infiltration, which is what a single-gene score is often measuring.
- The gain came from the combination, which is a stronger claim than a single panel performing well.
- Independence from clinical variables is the test that distinguishes a signature from a restatement of stage.
- A reversible marker that moves with recovery is a different and harder thing to build than a survival score.
Largest unresolved question
What counts as validation differs sharply across this literature. The immune aging clocks report internal and external validation on independent data, while the single-gene cancer markers are assessed within the same public repositories used to select them — so the reported AUCs and hazard ratios are not measuring the same property.
Common misconceptions
A gene that separates survival curves is a driver of the disease.
Both single-gene markers here correlate strongly with immune infiltration — ITGAL with CD8 T cells at rho = 0.732 and lower tumour purity. A gene expressed by infiltrating immune cells will track outcome without the tumour cells expressing it at all.
A high AUC means the signature is ready for clinical use.
Discrimination measured in the dataset that selected the genes is an upper bound, not an estimate. The studies here that report external validation on independent data are doing a different and harder thing than those reporting performance in the discovery repository.
Related
Claim ledger
What the evidence shows
Drawn from 5 studies in this library. Mix labels say which citation roles are present; they are not a strength score. Supports means evidence for a finding; Challenges means evidence against a stated position; Qualifies marks scope.
Both markers track immune infiltration, which is what a single-gene score is often measuring.
Single genes recur as prognostic markers across cancers. CDK2 showed highly significant prognostic value in lung adenocarcinoma (P = 5.8e-15) with higher tumour than normal expression and stage and survival links across a pan-cancer analysis; ITGAL discriminated gastric cancer with AUC 0.798 and correlated with CD8 T-cell infiltration (rho = 0.732).
The gain came from the combination, which is a stronger claim than a single panel performing well.
Combining data types improves classification over either alone. Integrated miRNA plus cytokine/chemokine profiling classified severe cases more accurately than either input, with severity-specific links separating patients whose cytokine levels were otherwise similar.
Independence from clinical variables is the test that distinguishes a signature from a restatement of stage.
Signatures can add information beyond standard clinical variables. A risk score built by integrating single-cell and bulk RNA sequencing separated overall survival independently of usual clinicopathology, with high-risk tumours showing more TP53 mutation, higher tumour mutational burden, PD-L1 upregulation and cell-cycle programmes.
A reversible marker that moves with recovery is a different and harder thing to build than a survival score.
Signatures can be built to track state rather than outcome. Single-cell immune aging clocks passed internal and external validation, showed monocyte age acceleration in COVID-19 patients that decreased during recovery, and marked heterogeneous vaccination responses by baseline interferon gene expression.
Debates
Tensions and limits
Some items are genuine disagreements on the same question. Others mark different assays, populations, or outcomes.
What counts as validation differs sharply across this literature. The immune aging clocks report internal and external validation on independent data, while the single-gene cancer markers are assessed within the same public repositories used to select them — so the reported AUCs and hazard ratios are not measuring the same property.
What counts as validation differs sharply across this literature. The immune aging clocks report internal and external validation on independent data, while the single-gene cancer markers are assessed within the same public repositories used to select them — so the reported AUCs and hazard ratios are not measuring the same property.
PaperFren reads this as a limit on how far one study travels — different assays, populations, or outcomes — not a forced fight between papers.
Timeline
How understanding moved
Study years are when the paper was published. Evidence edits are dated changes to this page's claims. Explanations are when PaperFren added a Discovery — not a claim that the science happened that day.
2026
Concept page published
Expression-Based Prognostic Signatures
Change log
What changed
Dated edits to this page's evidence: studies added or removed from a claim, claims added or withdrawn, and new explanations tagged here. Rewordings are not listed.
- Concept page published
Papers
5 studies in this library bear on Expression-Based Prognostic Signatures, ordered by citations.
- CDK2 anchors an immune/ceRNA prognosis model in LUAD
Across GEO/TCGA analyses, CDK2 is upregulated with strong LUAD survival association and is embedded in immune, Nomogram, ceRNA, and drug-IC50 networks.
- Can single-cell plus bulk RNA-seq predict LUAD survival?
From 8,170 lung cells and TCGA LUAD bulk data, a six-gene score (CP, GOLM1, CYP4B1, DAPK2, NFIX, FHL2) stratified survival (1-/3-/5-year AUC 0.67/0.67/0.64) and tracked TP53, TMB, and PD-L1.
- Can single-cell RNA clocks tell how old your immune system is?
sc-ImmuAging clocks trained on 1,081 healthy adults’ PBMCs showed monocyte age acceleration in COVID-19 that eased with recovery, and CD8+ T-cell rejuvenation after BCG mainly in people with high baseline interferon genes.
- Do miRNAs plus cytokines flag deadly COVID-19?
In 166 hospitalized COVID-19 patients, combining 171 microRNAomes with 25 cytokines discriminated severe disease better than either alone; CCL20, IL6, IL10, and miR-451a were key fatality correlates.
- Is ITGAL a gastric-cancer immune biomarker?
ITGAL mRNA is higher in 408 GC vs 211 normal samples (AUC 0.798), tracks CD8/macrophage infiltrates and PD1, and worse OS on KM plotter (HR 1.25 and 1.47) across 1,065 tumors.
Compare studies
Select 2–10 studies. Design and N are labels, not a ranking.
Nothing selected yet.
Questions
What is still open
What counts as validation differs sharply across this literature. The immune aging clocks report internal and external validation on independent data, while the single-gene cancer markers are assessed within the same public repositories used to select them — so the reported AUCs and hazard ratios are not measuring the same property.
Ask PaperFren about Expression-Based Prognostic Signatures
Study this conceptflashcards and short-answer questions
What is the first question to ask about any expression-based prognostic signature?
Whether the genes were selected and evaluated on the same data. The standard pipeline ranks genes by their correlation with survival and then reports that the resulting score predicts survival, which is guaranteed to look good. A signature that has been applied unchanged to an independent cohort — as the immune aging clocks were — is reporting a different quantity from one evaluated inside its discovery repository.
Why does a correlation with immune infiltration complicate a prognostic claim?
Because bulk expression mixes tumour and immune cells. ITGAL correlated with CD8 T-cell infiltration at rho = 0.732 and with lower tumour purity, so its apparent tumour expression partly reflects how many immune cells are in the sample. Since infiltration independently predicts outcome, the gene can track survival without being expressed by, or acting on, the tumour at all.
What makes a reversible marker harder to construct than a survival score?
It has to move in the right direction in the same people. The immune aging clocks showed monocyte age acceleration in COVID-19 patients that decreased during recovery — a within-person change tracking a known clinical trajectory, which overfitting to a cross-sectional outcome cannot produce. A survival score only needs to rank people once.