p-value and FDR

Question

[osamashehzad.786]

Suppose we are performing a gene expression experiment looking at the effect of a drug and we have found an interesting gene, geneX, that has a uncorrected T‐test p‐value of 10^−6. What is the corresponding Bonferroni adjusted p‐value if we started with 15,000 genes? Suppose I have a hypothesis that the drug affects genes in the glycolysis pathway (which geneX is a member of) and I will only consider those genes for hypothesis testing. What is the Bonferroni adjusted p‐value if there are 450 genes in the glycolysis pathway?

I have also computed the Benjamini‐Hochberg (BH) FDR correction for this experiment and it is 0.001 at the p‐value ≤ 10−6 threshold, when considering all 15,000 genes.Suppose that my hypothesis is correct and among the glycolysis pathway genes p‐values ≤ 10^−6 are 10 times more frequent the genome‐wide rate for all 15,000 genes.

What can you say about the BH FDR for the same threshold computed on the glycolysis restricted geneset.

Answer 1

[christopher medway]

Your Bonferroni corrected threshold for n=15000 will be; 0.05/15000 = 3.33e-06. While it is reasonable to reduce the multiple testing burden by selecting a subset of genes, this should have been decided upfront (not ad-hoc). Otherwise it becomes a fishing exercise. You may want to consider doing a formal "gene-set enrichment analysis" to see if your glycolysis pathway genes are significantly differentially expressed compared to other genes in the background set.

Answer 2

[Nicolas Rosewick]

You should correct on the number of test you did so 15000.

But if you want to test if significative genes (let say FDR < 0.1) are more enriched in the glycolysis pathway you should do a fisher exact test.