Hello, 

I have a question regarding the qvalues. I have a dataset of 240 metabolites and I did univariate tests, generating 240 p-values. I then calculated the qvalues from these pvalues, and got few qvalues that are actually lower than the pvalues (e.g. pvalue = 0.07; qvalue = 0.03). I understand that this is technically possible, but I'm not sure about the interpretation. In other forums I have read that this could be because the proportion of truly null p-values is not very accurately calculated when the number of tests is not large enough (in this case 240). Any advice??

Thanks very much, 

Andrea

You're correct that q-values will eventually be smaller than p-values when pi0 < 1.  See: https://github.com/StoreyLab/qvalue#frequently-asked-questions

There are finite sample results (meaning even for a small number of p-values) that show when the q-value estimates are conservative in Storey, Taylor, and Siegmund (2004) JRSS-B.

If you're concerned about the variance of your estimates, you could fix the lambda argument at a midrange number, such as lamba=0.5 or lambda=0.4.

Thanks very much for your detailed reply. Really useful.

Best wishes, 

Andrea