Binarizing significant/insignificant interactions

54 views
Skip to first unread message

Elena Del Pup

unread,
Dec 7, 2022, 2:51:07 PM12/7/22
to Fit-Hi-C
Dear all, 
I want to assign significant (intrachromosomal) interactions to 1 and insignificant to 0. Therefore, I am following the Benjamini Hochberg procedure to find the test with the largest p-value that is less than its Benjamini-Hochberg critical value, above which all other p-value-ranked tests are significant. 

I was wondering if it's ok if I process this way all chromosomes together or if I should rather process chromosomes individually.

Also, it would be a nice additional feature to directly have a column in the output flagging the significance. 

Thank you! 
Best, 
Elena 

Ferhat Ay

unread,
Dec 8, 2022, 1:50:09 PM12/8/22
to Fit-Hi-C
FitHiC is already doing the multiple testing correction using BH procedure. It does it genome-wide (right thing to do) if you run it genome-wide with all chromosomes. So you should be able to just use the  "q-value" column instead of p-value for filtering. No need to do anything else

Elena Del Pup

unread,
Dec 8, 2022, 2:17:47 PM12/8/22
to fit...@googlegroups.com

Dear Ferhat,

 

What cutoff should I use for the q value then?
Thank you.
Best regards,
Elena

--
You received this message because you are subscribed to a topic in the Google Groups "Fit-Hi-C" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/fithic/v8mwyPLTnWM/unsubscribe.
To unsubscribe from this group and all its topics, send an email to fithic+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fithic/aa76a021-5c59-4c94-9f38-a8ec77baede1n%40googlegroups.com.

Elena Del Pup

unread,
Dec 10, 2022, 1:11:34 PM12/10/22
to Fit-Hi-C
Dear Ferhat, 

I am confused by the way in which the Benjamini-Hochberg method works and how to find the significant tests. In the BH method (e.g., according to this guide) the signficance is defined by the all the tests (ranked by their p-value) above the test which has the largest p-value that is smaller that it's BH critical value. If the q-value you are reporting is the BH critical value, following this procedure makes me find a test with p-value 0.99999 and q-value 1. This means that all all p-values that are 0.9999.. or lower are significant. 

This does not seem quite right. A screenshot of this in the attachment. 

Also, it is not clear to me what is the FDR value that you are using to compute the q-value. 

Thank you! 
Best regards, 
Elena 
Screenshot 2022-12-10 at 10.07.09.png

Ferhat Ay

unread,
Dec 13, 2022, 2:24:10 AM12/13/22
to Fit-Hi-C
Your BH critical value can't be 0.99 etc. The standard thresholds people would use are 0.01 or 0.05. 
Please look into a tutorial or more accessible material to understand these concepts. Maybe this video helps.
 

Reply all
Reply to author
Forward
0 new messages