qqplots with RADmeth data
23 views
Skip to first unread message
Patrick McGowan
Jan 23, 2020, 7:23:28 PM1/23/20
to MethPipe and MethBase Users
Hi there,
I am posing this question in order to respond to a reviewer request for qqplots of data generated using RadMeth.
We cannot seem to find examples in the literature where qqplots for this type of data
are generated (except where different analysis methodologies are being compared).
The RadMeth output has many negative p values (i.e., -1) in the following conditions:
1. there is no read coverage
2. there is coverage but only in one condition
3. there is coverage but both conditions have 100 % methylation
4. there is coverage but no methylation counts for both groups.
In case of array data, situations 3 and 4 would still generate signals, and thus p values would be generated.
The question is whether it makes sense to do qq-plots with data from RadMeth (i.e. with the negative p values removed, and thus a restricted distribution of p values)? Any advice on how to proceed would be greatly appreciated.
1. there is no read coverage
2. there is coverage but only in one condition
3. there is coverage but both conditions have 100 % methylation
4. there is coverage but no methylation counts for both groups.
In case of array data, situations 3 and 4 would still generate signals, and thus p values would be generated.
The question is whether it makes sense to do qq-plots with data from RadMeth (i.e. with the negative p values removed, and thus a restricted distribution of p values)? Any advice on how to proceed would be greatly appreciated.
Many Thanks
Patrick
Ben Decato
Jan 23, 2020, 8:13:02 PM1/23/20
to meth...@googlegroups.com
Hi Patrick,
The negative p-values are just placeholders to say "the hypothesis wasn't tested" because it is trivial to see that there is no difference/not enough information in those cases, so removing them doesn't give you a restricted distribution of p-values, but rather the full set of hypothesis tests radmeth performed. It should be fine to produce a qq-plot based on these remaining tests.
One other thing I might recommend doing is permuting the values of the factor of interest a few times and rerunning radmeth to get a sense of what the null distribution of p-values in your data might look like. It might not be perfectly uniform and getting a sense of how deviant your observed p-values are from this permuted expectation would probably make the reply stronger.
Hope this helped,
Ben
The negative p-values are just placeholders to say "the hypothesis wasn't tested" because it is trivial to see that there is no difference/not enough information in those cases, so removing them doesn't give you a restricted distribution of p-values, but rather the full set of hypothesis tests radmeth performed. It should be fine to produce a qq-plot based on these remaining tests.
One other thing I might recommend doing is permuting the values of the factor of interest a few times and rerunning radmeth to get a sense of what the null distribution of p-values in your data might look like. It might not be perfectly uniform and getting a sense of how deviant your observed p-values are from this permuted expectation would probably make the reply stronger.
Hope this helped,
Ben
--
You received this message because you are subscribed to the Google Groups "MethPipe and MethBase Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methpipe+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/methpipe/94512827-4ebc-4baa-9dc3-47eb78f40536%40googlegroups.com.
--
Benjamin Decato, Ph.D.
Computational Biology Section,
Department of Biological Sciences,
University of Southern California
1050 Childs Way, RRI 408M
Los Angeles, CA 90089-2910
Benjamin Decato, Ph.D.
Computational Biology Section,
Department of Biological Sciences,
University of Southern California
1050 Childs Way, RRI 408M
Los Angeles, CA 90089-2910
Reply all
Reply to author
Forward
0 new messages