Big FDR values for 42 sample

[AndreaPeC]

Hello,
Thanks so much for this tool!!!

I’m having some issues understanding why I’m getting such big FDR values,
I’m currently working with 42 samples, 2 groups and 1000 permutations and I’m getting that go from 0.5 to 0.99.
At the beginning I thought it may be because of the number of permutations but I have also tested for 10, 20 and 100, and the FDR values stay on the same range, only the number of DMR changes.

Hope you can help me understand why this is happening?

Thank you!!!

[kdkorthauer]

Hi @AndreaPeC,

Thanks for reaching out. To understand what's happening here, I would need to know a lot more about your study. To start:

• What do the two groups represent, and do you expect these to have widespread DNAme differences?
• What other factors may differ from sample to sample, and have you considered adjusting for them in some way (e.g. sex)?
• How do the plots of, say, the top 10 DMRs look? Do you see significant separation between the two groups?
• What beta value cutoff are you using (cutoff), and approximately how many candidate regions do you find with that cutoff? Approximately how many null candidate regions are found per permutation?

[AndreaPeC]

Thank you so much for your answer,

I’m working with normal weight (NP) overweight/obese (OWOB) children, so I do expect to have methylation differences. About the covariates I hadn’t consider adjusting by sex but with your comment I did try it with 10 permutations and a cutoff of 0.05 but the number of significant regions (p < 0.05) became smaller, 10 regions, and the FDR values didn’t change, they’re still around 0.99. When I see the plots there is not much difference between the groups.

About the candidates, I got 1958 candidates detected, which is also the total regions I get, before filtering by p value, and in average I got from 800 to 2000 null candidates per permutation.

[kdkorthauer]

Hi AndreaPeC, thanks for the additional information. It sounds like if the regions that are detected don't visually look like convincing DMRs, I wouldn't expect them to have smaller FDR values.