50000 significant p values

Question

[chowdhury]

I have ADNI microarray dataset (https://ida.loni.usc.edu/pages/access/geneticData.jsp#207). The paper says it is normalized data. I have used R -limma package to do the expression analysis among healthy controls and patients. It has around 50000 probe set and sample size of 700 (equal amount of control and patients). After the analysis, all 50000 probe ids showed high significance level (<0.01). Is it possible?

[kloetzl]

What p-value correction did you apply (if any)?

[WouterDeCoster]

Is this normal?

No.

But you'll need to add more information to this question in order to get accurate troubleshooting, such as which technology, which experimental setup, which packages, which code,...

[chowdhury]

I have ADNI microarray dataset (https://ida.loni.usc.edu/pages/access/geneticData.jsp#207). The paper says it is normalized data. I have used R -limma package to do the expression analysis among healthy controls and patients. It has around 50000 probe set and sample size of 700 (equal amount of control and patients). After the analysis, all 50000 probe ids showed high significance level (<0.01). Is it possible

[WouterDeCoster]

Is it possible

No.

all 50000 probe ids showed high significance level (<0.01)

Are all transcripts up- or downregulated or both directions?

The paper

Which paper? Be specific.

And please use ADD REPLY to answer to earlier comments, as such this thread remains logically structured and easy to follow.

[mastal511]

@chowdhury Are you sure you are dealing with gene expression data? The link you have given says geneticData.

The only information I can find about ADNI mentions only genotyping arrays. http://adni.loni.usc.edu/methods/genetic-data-methods/

and the following 2010 paper about the Alzheimer's Disease Neuroimaging Initiative also mentions only Illumina genotyping arrays.

/https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2868595

Answer 1

[kloetzl]

Judging from your comment you did not apply any p-value correction. As you are doing multiple tests, you have to change your acceptance level from 0.01 to 0.01/50000 in the simplest case (Bonferroni). That then is the true alpha-value you have to compare with.

[WouterDeCoster]

Bonferonni is not appropriate (too stringent) for gene expression analysis. In addition I would expect Limma to perform multiple testing correction.

[boczniak767]

I agree, False Discovery Rate (FDR) in limma is appropriate.