What you get
A ranking of your candidate reference genes by expression stability, from your own samples, by two methods side by side. geNorm gives each gene its M value and says how many genes are enough and which ones, where the pairwise variation V drops below 0.15. NormFinder ranks the same genes by their variation within and between your sample groups.
A comparison table sets the two rankings next to each other and marks the genes where the methods disagree. Save the method you settle on as an analysis of the experiment: it reopens exactly as saved, with the data it was computed on.
Is it the geNorm you would cite?
Yes, with NormFinder beside it; Clarida’s defaults come from co-authors of geNorm. The methods are the published ones, applied to your Cq values after the experiment’s own checks.
| geNorm | M per gene, the pairwise variation V for every step from two genes up, and the cut-off of 0.15 (Vandesompele et al., 2002) |
|---|---|
| NormFinder | Variation within and between your sample groups (Andersen et al., 2004); with four or more candidates, the best pair |
| When the two disagree | The gene is marked for you to look at; there is no combined rank |
| The free geNorm and NormFinder tools | Enough to rank a Cq table you already have, with the same engine and no account. The experiment adds the panel’s design, the run, and the saved ranking with its data |
How it works in Clarida
01In the app: Design › Sample Prep
Samples that cover every condition
A reference gene is only stable for the conditions you tested it on, so the samples should span each treatment, tissue and time point you will compare later. Give them a property such as treatment when you paste them; NormFinder groups the samples by it.
- Properties: text, a list, a number, yes or no, a date; any of them can group the samples
In full on the Design page
02In the app: Design › Assay Prep
Your candidates, marked as reference genes
The Reference gene finder ranks exactly the targets typed as reference genes, and needs at least three. With fewer, it says so and offers to treat every target as a candidate, which you choose yourself; it never happens on its own.
- Default type: gene of interest, so mark each candidate
In full on the Design page
03In the app: Analyze › Cq processing › Efficiency
Each candidate at its own efficiency
The ranking uses each target’s efficiency as Clarida resolved it for the experiment: a standard curve, the library value or the default. An efficiency outside 1.5 to 2.5 is replaced by 2, and the Reference gene finder names the target.
- Default: automatic, this experiment’s standard curve first, then your library, then 2
In full on the Analyze page
04In the app: Analyze › Reference gene finder
geNorm and NormFinder on the same data
Choose a grouping, leave out a gene or a sample with one click (it stays visible, to include again), and press Analyze. Both methods run on one Cq per sample and target, left after your exclusions. A short data check lists any sample or gene left out for missing values. Save keeps the method on screen; a saved one opens read-only, and an editable copy lets you try other choices.
- Group samples by: none by default, or any property you gave them (NormFinder uses it; geNorm does not)
The free tool: geNorm, and how the method works
Why it is done this way
No gene is stably expressed in every tissue and condition, and normalizing to a single unvalidated one can lead to large errors in part of your samples. Vandesompele et al. (2002) showed this across ten common reference genes and proposed the geometric mean of several validated ones; geNorm ranks candidates by M, and V decides how many are enough.
geNorm can rank co-regulated genes as stable, because they vary together. NormFinder estimates the variation within each group of samples and the shift between them, so a gene that moves with your treatment is caught (Andersen et al., 2004). Where both agree, you can be confident; where they disagree, the table points to the genes to look at.
Validation is done once per experimental system. In the experiments that follow, Clarida normalizes on the geometric mean of the validated genes (Hellemans et al., 2007) and reports their M and coefficient of variation again, so you see the genes still behave.
What it doesn’t do
| Setting your reference genes from the ranking | You mark the chosen genes yourself, in the experiment and in your library |
|---|---|
| BestKeeper, the comparative ΔCt method, a combined rank | geNorm and NormFinder only, side by side |
| Normalization and statistics | In a gene expression experiment: this one validates genes |
| Genes not measured on the same samples | Left out until the data are complete; the data check lists each |
Frequently asked questions
Is GAPDH a safe housekeeping gene?
Only if it is stable in your samples, which is what the ranking shows. No gene is stable in every tissue and treatment, and a common one such as GAPDH can shift with yours.
How many candidate genes and samples do you need?
Either method needs at least 3 candidates and 3 samples. NormFinder also needs 2 samples in each group. Its authors recommend 5 to 10 candidates and 8 samples per group or more, and the app warns you below that.
Do you need sample groups?
Not for geNorm, which ignores them. NormFinder can run without groups, but it is most useful when they follow the condition you will compare.
How do the chosen genes reach your next experiment?
Save the candidates to your library with the chosen ones typed as reference genes, and take them into your next gene expression experiment. Normalization there uses that type and checks their stability again.