# Choose reference genes in the example

> Three candidate reference genes, one unstable. Find it, leave it out, and see the treatment effect it was hiding.

Source: https://clarida.bio/help/example-reference-genes
Updated: 2026-11-03

[Help](https://clarida.bio/help)

Three candidate reference genes, one unstable. Find it, leave it out, and see the treatment effect it was hiding.

By [Ward Hellemans](https://clarida.bio/help/example-reference-genes#about-the-authors)

November 3, 2026 4 min read

Every new Clarida workspace comes with this example: a treatment study with two independent groups, 16 samples, 8 untreated and 8 treated. Three genes of interest were measured together with three candidate reference genes, in duplicate, on three plates, with a dilution series of five standards and a no-template control. Sample and gene names are illustrative.

Normalization divides every measurement by a factor computed from the reference genes, so a result is only as good as those genes. In this example, all three candidates start included. Step 2 shows whether they should be.

It is your workspace's copy: colleagues in the workspace see the same one. You can explore every analysis step and change how results are shown; the measured design (samples, targets, plates) is locked so the example stays intact.

![NRQ bar chart of the gene Palm, grouped by Treatment: about 1.9 in the untreated group (No) and about 0.5 in the treated group (Yes)](https://clarida.bio/images/help/refgenes-hero.png)

Figure 1: Where this guide ends: Palm, about 3.6 times lower in the treated group, once the reference genes can be trusted.

## How to

### 1. Open the example and read its goal

Open the unpaired comparison example: a card on your home page or, once you have an experiment of your own, in the experiment list with the **Example** badge. The experiment opens on **Define**: the **Background** describes the study, and the **Goal** says what it sets out to answer.

![Define: the example's name, the Background describing the two-group study, and the Goal: select stable reference genes, then determine whether the treatment changes expression](https://clarida.bio/images/help/refgenes-define.png)

Figure 2: Define: the Background and the two-part Goal.

**What to notice:** the Goal has two parts, and the order matters: first choose the reference genes, then look at the genes of interest. Scroll down inside the Background box for the places worth visiting. Above the experiment, the **Demo experiment** banner says what you can do: analysis, annotations and reporting all work; structural edits are locked.

### 2. Spot the unstable candidate

Go to **Analyze → Cq processing → Normalization**. The table lists the three candidate reference genes with two stability measures: **M** (the geNorm stability value) and **CV (%)**, the coefficient of variation. Lower is more stable for both. The chart below shows the normalization factor of every sample.

![Normalization with three reference genes: the warning Reference genes not stable enough (mean M 2.696, mean CV 110.3%); Flexible has M 3.946 and CV 147.5%, Stable and Nonregulated about 2.07 and 90%; normalization factors range from about 0.1 to 6.4](https://clarida.bio/images/help/refgenes-normalization-before.png)

Figure 3: Normalization with all three candidates: the warning, and Flexible's M and CV.

**What to notice:** Clarida warns **Reference genes not stable enough**: the mean M is 2.70, above the 0.5 cut-off, and the mean CV is 110%, against 20%. One candidate stands out: **Flexible**, with an M of 3.95 and a CV of 148%. The other two look poor as well, but a gene's M is measured against the other reference genes, so one unstable gene drags every value up.

### 3. Leave it out

In the **Include / Exclude** column, click the cross on the **Flexible** row. Clarida recalculates the stability of the remaining genes and every normalization factor straight away.

![Normalization after excluding Flexible: Normalizing on 2 stable reference genes, Stable and Nonregulated with M 0.194 and CV 6.7%, all checks green; the normalization factors now range from about 0.1 to 2.2](https://clarida.bio/images/help/refgenes-normalization-after.png)

Figure 4: Normalization without Flexible: two stable reference genes, all checks green.

**What to notice:** the warning turns into **Normalizing on 2 stable reference genes**. Without Flexible, the mean M drops from 2.70 to 0.19 and the mean CV from 110% to 6.7%: Stable and Nonregulated agree closely. The normalization factors change too. Sample11 and Sample14 seemed to have about six times the average input; with two stable reference genes it is about twice. Flexible stays in the table, so you can include it again at any time.

### 4. Compare the two groups

Go to **Exploration**. The **NRQ bar chart** shows the first gene. Under **Grouping**, set **Property 1** to **Treatment** and choose **Plot group average**. Then use the arrows next to **Target** to step through the three genes of interest.

![Exploration: Data health with 2 replicate sets exceed threshold, 1 target with borderline efficiency, normalization on 2 stable reference genes and 4 sample and assay QC alerts; the NRQ bar chart of Palm grouped by Treatment, plotted as group averages; untreated (No) about 1.9 and treated (Yes) about 0.5](https://clarida.bio/images/help/refgenes-exploration.png)

Figure 5: Exploration: Data health, and Palm's group averages by treatment.

**What to notice:** for **Palm**, the treated group is about 3.6 times lower than the untreated group (1.9 against 0.5; the **Ratio (A/B)** column shows 3.584). **Duvel** and **Leffe** barely differ between the groups.

The **Data health** summary at the top collects the checks this result rests on, and three are orange:

- **Sample & assay QC**, **4 alerts**: the unequal input from step 3. Sample01, Sample11 and Sample13 have normalization factors outside the 0.5 to 2 range, and Sample13, with the least input, also has a high average Cq. Normalization corrects for this.
- **2 replicate sets exceed threshold**: in Sample05 for Palm and in Standard5 for Leffe, one of the two replicates is almost 4 cycles away from the other.
- **1 target with borderline efficiency**: Leffe, at 1.74, pulled down by that deviating Standard5 replicate.

The alerts ask you to take a look; they don't mean the analysis failed. The two deviating replicates are a good place to start; see below.

## Try this

With the result in view, change one thing and watch the analysis follow.

- **Bring Flexible back.** Include it again in **Normalization**, then return to the Palm chart. How much of the difference between the groups survives an unreliable reference gene?
- **Exclude the deviating replicates.** Under **Cq processing → Exclusion**, exclude well A10 (Sample05, Palm) and well H10 (Standard5, Leffe). Watch the replicate and efficiency checks turn green, then return to the Palm chart. Does the difference between the groups change?
- **Look at every sample.** Switch from **Plot group average** to **Plot individual samples** to see how much the eight samples within each group vary.
- **Change the scale.** Set the **Y-axis scale** to **Log**, so a twofold increase and a twofold decrease look the same size.

## What you just did

- Checked the stability of three candidate reference genes.
- Left out the one that could not be trusted, and saw the others turn stable.
- Saw how the normalization factors change with the reference genes.
- Compared the treatment groups for each gene of interest.

To do the same on your own run, [Importing qPCR run files and Cq (Ct) tables](https://clarida.bio/methods/importing-instrument-data) says which exports Clarida reads and how they are matched.

## About the author

- Ward Hellemans

  Co-founder, Clarida

  Runs the product and the company side of Clarida.

## Stuck on a reference gene?

A founder answers within a working day. Say which step, or which instrument wrote the file you import.

[Start free](https://app.clarida.bio/login) · [Talk to a founder](https://clarida.bio/talk-to-us?topic=help)
