Calculation of multiple matrix FDR

[Bharata Kalbuaji]

Hello,

I want to ask about the FDR in the FIMO. Suppose I have a promoter sequence of length 1000nt positive strand. If I check 100 PWM, I would have several matches for each PWM. I want to ask whether the multiple hypothesis testing is grouped by each matrices?

For example, matrix 1 will have 20 match. We apply the FDR (or q-value) for only those 20 matches and we will do the same for the rest of PWM. So, each FDR is applied to only 1 matrix result. On the other hand, does the FIMO calculate all p-value and then calculate the FDR (or q-value) from all generated p-value regardless which matrix gives the result? Is there any difference from both method to calculate FDR? Thanks.

[CharlesEGrant]

The distribution of p-values is different for each PWM, so q-values are computed separately for each motif. The FDR should be considered separately for each motif. Note thoug,h that the q-values provided by FIMO are compensating for applying the p-value threshold for a single motif at each point in your sequences. In a previous posting you noted that you are scanning with 4000 motifs, so you are introducing a second, distinct, source of multiple tests issues, which are not considered by FIMO. You may want to consider applying a Bonferroni correction to the FIMO q-values, say multiplying them by 1e4 or 1e3 (roughly 1/4000) . 

[Bharata Kalbuaji]

I see thank you very much. Now it becomes clear to me.