Fwd: RADMeth negative p-values

[Andrew D. Smith]

Hi Patrick.

Sorry for the slow response. I'm forwarding this to the google group in the hopes that someone else might be faster at answering this than me.

Cheers,
Andrew

Begin forwarded message:

> Resent-From: <andr...@usc.edu>
> From: Patrick McGowan <patrick...@utoronto.ca>
> Subject: RADMeth negative p-values
> Date: June 20, 2018 at 4:30:29 PM PDT
> To: "andr...@usc.edu" <andr...@usc.edu>
> Cc: Aya Sasaki <aya.s...@utoronto.ca>
>
> Dear Dr. Smith,
>
> We have been implementing RADMeth within MethPipe for analysis of RRBS data and have noticed that it generates p-values ranging from -1 to +1 for case/control comparisons.
>
> The presence of negative p-values is puzzling to us. I was wondering whether this is an expected output and if so how to interpret them?
>
> Thank you in advance for any assistance you can provide.
>
> Best wishes,
> Patrick
> -----------------------------------
> Patrick O. McGowan, PhD
> Associate Professor
> Biological Sciences, Scarborough
> Cell and Systems Biology, Psychology, Physiology
> University of Toronto
> SW-548, 1265 Military Trail, Toronto ON M1C1A4 Canada
> Tel 416.208.5153 | Fax 416.287.7676
> patrick...@utoronto.ca
> https://urldefense.proofpoint.com/v2/url?u=http-3A__www.utsc.utoronto.ca_-7Epmcgowan_&d=DwIFAg&c=clK7kQUTWtAVEOVIgvi0NU5BOUHhpN0H8p7CSfnc_gI&r=2ti3K9K1XjkPszCPHB82Aw&m=vgoRcHE4Jy0M0AG9th_cSgL47-QUq4bgAfnHkXb4aWI&s=k_4pyBnWW_Q6ozv3dbL5fnk2xj_toc_zEiKyS_i2WPU&e=
>
>
>
>

[Ben Decato]

Hi Patrick,

Are there any p-values greater than -1 but less than 0? If so, then that's probably a bug. A p-value of -1 was intentional, and indicates that there was either:

1) no coverage in any of the samples for that site
2) all methylation levels at that site are the same across all samples
3) an error occurred in the fitting algorithm (i.e. p-value is nan or -nan)

In these cases, no test is performed or likelihood computed. We thought it would be more informative than a valid p-value because they can be excluded prior to multiple testing correction. I'll make sure a note about this is added to the methpipe manual, and will probably change these codes to -1, -2, and -3 for more specificity in our development branch.

Hope this helps!

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+unsubscribe@googlegroups.com.
For more options, visit https://urldefense.proofpoint.com/v2/url?u=https-3A__groups.google.com_d_optout&d=DwIFaQ&c=clK7kQUTWtAVEOVIgvi0NU5BOUHhpN0H8p7CSfnc_gI&r=SpXCeWczJOWPYEneTspxwg&m=4lbLkEm9AxOcPOnHncvDuiWySDNdIMOfxkTuDW5yyLc&s=oqM9H-vqMduc3x8pQhe9OWwaf2NVT8gpeuoqByQBhR0&e=.

--

-- 
Benjamin Decato
PhD Candidate
Computational Biology Section,
Department of Biological Sciences,
University of Southern California
1050 Childs Way, RRI 408M
Los Angeles, CA 90089-2910

[Andrew D. Smith]

On Jun 26, 2018, at 12:21 PM, Ben Decato <dec...@usc.edu> wrote:

>
> Hi Patrick,
>
> Are there any p-values greater than -1 but less than 0? If so, then that's probably a bug. A p-value of -1 was intentional, and indicates that there was either:
>
> 1) no coverage in any of the samples for that site
> 2) all methylation levels at that site are the same across all samples

I should note that no variation at all makes it difficult to fit certain parameters, which in one sense corresponds to case (3) below, but we might be better off just reporting a p-value of 1.0.

The problem is that often these 3 cases are the same, and at some point in development were likely separated as a means of earlier identifying case (3), because no coverage (i.e. (1)) will imply (2), and in our procedure, (2) will imply (3). And the most common source of (3) before checking the presence of (1) or (2) was probably just (1).

But at this point, we should report these as distinct values and make sure they are documented.

I'd be interested in hearing any opinions in favor of *not* reporting a 1.0 for case (2). I vaguely recall a reason for that choice. Possibly having to do with skewing the overall distribution of p-values, which is important for subsequent FDR calculations. But even that could be handled.

> To unsubscribe from this group and stop receiving emails from it, send an email to methpipe+u...@googlegroups.com.

> For more options, visit https://urldefense.proofpoint.com/v2/url?u=https-3A__groups.google.com_d_optout&d=DwIFaQ&c=clK7kQUTWtAVEOVIgvi0NU5BOUHhpN0H8p7CSfnc_gI&r=SpXCeWczJOWPYEneTspxwg&m=4lbLkEm9AxOcPOnHncvDuiWySDNdIMOfxkTuDW5yyLc&s=oqM9H-vqMduc3x8pQhe9OWwaf2NVT8gpeuoqByQBhR0&e=.
>
>
>
> --
> --
> Benjamin Decato
> PhD Candidate
> Computational Biology Section,
> Department of Biological Sciences,
> University of Southern California
> 1050 Childs Way, RRI 408M
> Los Angeles, CA 90089-2910
>

> --
> 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.
> For more options, visit https://groups.google.com/d/optout.