shall I select genes based on p-value alone if FDR is not significant?

Question

[muthulaxmi.m]

In general, differential gene expression analysis from microarray profile uses p-value along with multiple corrections to get adjusted p-value. I have a doubt that, if none of the genes satisfy the criteria of FDR<0.05, shall I select significant genes based on p-value<0.05 criteria alone?

Since I could not find any literature supporting the filtering of genes based on p-value alone, can anyone guide me in selecting the significant genes. Is this case of FDR>0.05 is allowed in microarray data analysis?

[Jean-Karim Heriche]

It's preferable to use adjusted p-value to account for multiple testing. The threshold is arbitrary. In the case of FDR you can think of it as the fraction of false positives so if you pick genes at a FDR < 0.1 then 10% of the genes you've selected will be false positives. How you set the threshold depends on how costly the errors are going to be, e.g. is it worth doing follow up experiments on 100 genes knowing that 10 are false positives ? In general if you don't want to abide by the result of a statistical test then don't do it in the first place and if you don't care about statistical significance then just go ahead and pick genes based on whatever criteria suits you.

Answer 1

[Devon Ryan]

If none of the genes are significant given your desired cut off then don't select anything. It is, after all, perfectly normal for experiments to show no change.