The t-test can only be used if your data doesn't violate any of the t-test assumptions (equal variance, normal distribution etc.). Although the t-test is (erronously) commonly used to measure differences between two samples, you have to know what the distribution of your data is, to make sure that you're not using the wrong test for your analysis. Especially if you have very small sample sizes (for example n = 3 per group), a so-called non-parametric test such as the Mann-Whitney U (aka Wilcoxon Rank Sum) is usually recommended. There are issues with using the Mann-Whitney U test, especially in the context of gene expression data, as outliers may affect your results, but especially in the context of small sample sizes, you'll usually be quite safe using the Mann-Whitney U test. A third and final option, is to log-transform your data and then run a t-test. This may be necessary if you have larger sample sizes (for example n = 50 per group), but your data doesn't fit a normal distribution (as is often the case with gene expression data).

You can chose these tests in Prism (most commonly used by scientists) as the others have recommended. Just put your data into columns (chose the colomun layout) and then click "analyze". Chose t-tests and other non-parametric tests. Then you can choose paired (for example gene one pre/post exposure, if it's the same cells/animal etc.) and non-parametric and then the Mann-Whitney U. You can do the same thing for the t-test. But please bear in mind although the "t-test" is the test name you most commonly encounter, its not always right to use.

Finally, I recommend you visit a stats course for researchers, and I don't mean this in an derogatory sense, but simply because it'll help you chose the right tests for your data, which is sometimes relevant, when trying to get thos p-values < 0.05.

HTH,

Oliver

Hi Richard

A simple Student t-test would show the statistical difference between the expression of your genes.

You an use GraphPad Prism software. All you have to do is feed the expression values (i guess you calculated for each gene based on Ct value) for each gene in column graph. Software will plot graph for you and provide option to perform various tests. Choose Student's t-test and its done.

Hope it helps!

Hi, Richard

This link below can be useful for you:

http://bioinformatics.gene-quantification.info/

It contains some information for qPCR data analysis. I hope you can find some answers.

Best regards,

student's t test as advised earlier using graph pad prism and you can choose paired t test (two tailed) if you want to compare gene 1 expression (untreated to treated). If you want to compare gene1 vs gene2 (say treated condition) use unpaired t test two tailed. good luck!

Thanks for all of the answers. I will look do the t-test and refer to the posted link for further understanding. Thanks again!

For any kind of statistical analysis go for "spss statistical analysis" software...easy to use and freely available

You could do a t-test on the differences between untreated and treated expression for the 2 genes (unpaired 2-tailed t-test on (gene1 treated - gene1 untreated) vs (gene2 treated - gene2 untreated))

The t-test can only be used if your data doesn't violate any of the t-test assumptions (equal variance, normal distribution etc.). Although the t-test is (erronously) commonly used to measure differences between two samples, you have to know what the distribution of your data is, to make sure that you're not using the wrong test for your analysis. Especially if you have very small sample sizes (for example n = 3 per group), a so-called non-parametric test such as the Mann-Whitney U (aka Wilcoxon Rank Sum) is usually recommended. There are issues with using the Mann-Whitney U test, especially in the context of gene expression data, as outliers may affect your results, but especially in the context of small sample sizes, you'll usually be quite safe using the Mann-Whitney U test. A third and final option, is to log-transform your data and then run a t-test. This may be necessary if you have larger sample sizes (for example n = 50 per group), but your data doesn't fit a normal distribution (as is often the case with gene expression data).

You can chose these tests in Prism (most commonly used by scientists) as the others have recommended. Just put your data into columns (chose the colomun layout) and then click "analyze". Chose t-tests and other non-parametric tests. Then you can choose paired (for example gene one pre/post exposure, if it's the same cells/animal etc.) and non-parametric and then the Mann-Whitney U. You can do the same thing for the t-test. But please bear in mind although the "t-test" is the test name you most commonly encounter, its not always right to use.

Finally, I recommend you visit a stats course for researchers, and I don't mean this in an derogatory sense, but simply because it'll help you chose the right tests for your data, which is sometimes relevant, when trying to get thos p-values < 0.05.

HTH,

Oliver

I'm agree with Oliver. I think the best, if you have small smples sizes as I can imagine, is to use non parametric test as the Mann-Whitney U. Furthermore, if you have housekeeping genes expression too, I recommend using geNORM software (Vandesomple, et al., Genome Biology, 2002. Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes) to search for best housekeeping gene()s and best normalization.

Best Regards,

Yvan

If you are familiar with R and bioconductor you can take a look at the library ddCt. It was specifically developed to perform statistcal analysis on of Quantitative Real-Time PCR (qRT-PCR).

http://www.bioconductor.org/packages/release/bioc/html/ddCt.html

Thanks for all the input! When do the statistical analysis, what values should I be using? Is it best to use the dCT values?

Also, I just got an email back for a qPCR specialist from BioRad and he said that since Gene 1 treated was significantly different from Gene 1 untreated, and Gene 2 treated had no significant difference compared to Gene 2 untreated, that this shows differential expression of the two genes, and further statistical tests are unnecessary. What are your thoughts on that?

You wrote : "I want to show that the difference between Gene 1 and Gene 2 expression in response to the treatment is significant". In this goal, if you find signifiant difference in expression for gene 1 and not for gene 2 bête en treated and untreated, thus, that's true that you have a signifiant behaviour difference between the 2 genes....

Yeah, I thought that before as well. I just thought that I might have to do a statistical test comparing the two genes together (ie. Gene 1 treated vs. Gene 2 treated), instead/as well as comparing the expression of the individual genes when treated and untreated (ie. Gene 1 treated vs. Gene 1 untreated, compared to Gene 2 treated vs. Gene 2 untreated). Do you think an addition statistical test is needed for the first example I just gave (ie. Gene 1 treated vs. Gene 2 treated). And for that test, I think it would just be a simple t-test, because I am just comparing two groups and the data is normalized. Is my line of thinking correct?

yap... you can just perform a simple t-test for two groups: gene1 vs gene2 (treated). I think that would suffice.

you could use REST software that is designed for qPCR, you could download free http://rest.gene-quantification.info/

You'll be looking at delta-delta-delta-CT across groups I believe, so:

((g1 treated  delta CT - g1 untreated delta CT) - (g2 treated delta CT - g2 untreated delta CT)) / error term

@Oliver

I wonder why you think that Mann-Whitney U test in appropriate for small data sizes (e.g. 3 replicates). This issue is clearly explained in GraphPad Prism statistics guide: "If you have small samples, the Mann-Whitney test has little power. In fact, if the total sample size is seven or less, the Mann-Whitney test will always give a P value greater than 0.05 no matter how much the groups differ."

http://www.graphpad.com/guides/prism/6/statistics/index.htm?stat_the_power_of_nonparametric_tes.htm

http://www.graphpad.com/guides/prism/6/statistics/index.htm

it is better to use a non-parametric test (e.g Mann-Whitney) for testing significance of qRT-PCR results. CTs are usually not normally distributed so u can't carry out a parametric test like t-test.

u can use Shapiro-Wilk test to test normality first then according the result u can decide which one to use

How did you calculate p-value for fold-change?