correcting for multiple testing?

[aprice26]

I was just wondering if the Audic and Claverie Test p-value results were already corrected for multiple testing in the .tab output, and if not, what would you recommend as the best way to do that?

Thanks,

Amanda

[dg520]

Hi Amanda,

The p value of Audic and Claverie test in IRFinder is calculated for each intron independently, thus we indeed need a multi-test correction afterwards to minimize false positives. It's an open topic what the best correction is in the genomics field. Bonferroni is used in GWAS study while FDR (BH) is applied in gene expression, in common. It all depends on what distribution of true null hypothesis we believe in. I would say that using either of these will be widely accepted.

Best, Dadi

[aprice26]

How are the introns reported determined? The Audic and Claverie results contain a few thousand introns, but the IRFinder-IR-nondir.txt files include many more introns. I'm wondering how to correct for FDR when not all p values are reported (assuming all the IRFinder-IR-nondir.txt introns are tested). I can use the p.adjust() function in R, but it will assume the unreported p values are all equal to 1).

Thanks,

Amanda

[dg520]

Hi Amanda,

Good question and my bad not making it clear. The built-in A-C test of IRFinder only applies to "high confident" introns, by filtering out introns that are not suitable for differential IR analysis. For example in a KO-vs-WT design, if an intron is highly expressed in WT while both this intron and its flanking exons are not well expressed in KO, it's hard to determine whether this is a differential IR case or a case with some technical issue during sequencing.

The filter works as following steps:

1. If in both conditions, an annotated intron is supported at the exact splice junction in the sequencing data (marked as "-" in the last column of original IR result of each condition), the fake p value is initiated as 99; Otherwise, the intron will not be tested and reported
2. A secondary filter check if the number of reads supporting the splicing of this intron is no less than 10 in either condition. If only one condition passed this filter, the initiated 99 will be updated to 33. If both condition passed, the intron will be send to A-C test and reported with a real p value.

So you can set all "33" and "99" to "1", and apply p.adj() in R. You can find the explicit filter setting inside the "bin/analysisWithNoReplicates.pl" (near the bottom) if you are comfortable with perl.

Best, Dadi

[aprice26]

Thanks for the explanation! Just to be clear, I should set all introns from the IRFinder-IR-nondir.txt result that are not assessed by the A-C test and output in the .tab file to 1?

[dg520]

Yes Amanda, that's what you wanna do.

Best, Dadi