2

I am working with an experiment on several genes (#41350) based on two factors: Lcells (levels: "pos", "neg"), microbe(levels: "CR", "GF"). I have implemented ANOVA test (aov) for each individual gene separately comparing Lcells*microbe. The result is different p-values of Lcells, microbe and the interaction for each individual gene (a matrix of 41350*3). Now, how should I adjust p-values? What is the vector I should feed into p.adjust()? Shall I adjust all Lcells p-values of all genes together, then microbe p-values of all genes separately and ...?

Thanks for your help!

Sven Hohenstein
  • 6,833
Rozita Akrami
  • 71

1 Answers1

1

What you suggest is what is usually done. You typically adjust all the p-values for each factor and their interaction separately.

I do not know what test you are using, but for example, this method is what is used inherently in the functionality of the limma package from the Bioconductor project.

  • but that's strange, since before adjusting p-values, for each individual factor I have more than 1500 significants. but after adjusting no significant p-value would remain?! – Rozita Akrami Sep 29 '13 at 06:55
  • Well, this often happens, especially if the conditions you are comparing are biologically only slightly different, or if you have a low number of replicates. –  Sep 29 '13 at 07:17
  • What method of adjustment are you using? –  Sep 29 '13 at 07:17
  • I am using 'fdr' method – Rozita Akrami Sep 29 '13 at 07:41
  • ANd I should say that yes I have low number of replicates. only 3 replicates! – Rozita Akrami Sep 29 '13 at 07:42
  • 1
    Is your aim to validate the findings, also? If so, the top n genes are still the ones that you would like to use for validation, because FDR does not change the ranking of the genes. In addition, not having any significant p-values left does not (necessarily) mean that you do not have any differences between the conditions, you just have a low power to detect these differences. –  Sep 29 '13 at 07:49
  • Depending on the test that you are using, you might consider filtering your data set using e.g. standard deviation of low expression before running the tests and adjusting the p-values. That is, if haven't already. This typically leaves you with more DEG genes. –  Sep 29 '13 at 07:51