Page 1 of 1

[SOLVED] Zeros in mlinfo file results in NAN beta and sebeta

Posted: Tue Jul 10, 2012 8:04 pm
by vforgetta
Hi,

In the ProbABEL manual it states that the values for Freq1, MAF, Quality, and Rsq in the MLINFO file do not matter:

"Actually, for ProbABEL, it does not matter what is written in this le
{ this information is just brought forward to the output."

I found that in some regard they do. If I place zeros for all these values, like so:

Code: Select all

SNP Al1 Al2 Freq1 MAF Quality Rsq
7-39992470 A G 0 0 0 0
7-39992529 C T 0 0 0 0
7-39992552 A T 0 0 0 0
7-39992766 C T 0 0 0 0
7-39992930 A G 0 0 0 0
7-39992946 G C 0 0 0 0
7-39992986 G A 0 0 0 0
7-39993073 C G 0 0 0 0
7-39993085 G A 0 0 0 0
7-39993183 A G 0 0 0 0


I get a results like this:

Code: Select all

name A1 A2 Freq1 MAF Quality Rsq n Mean_predictor_allele position beta_SNP_add sebeta_SNP_add
7-39992470 A G 0 0 0 0 3172 7.69231e-05 7-39992529 nan nan
7-39992529 C T 0 0 0 0 3172 9.45776e-07 7-39992552 nan nan
7-39992552 A T 0 0 0 0 3172 0.000214061 7-39992766 nan nan
7-39992766 C T 0 0 0 0 3172 0.000245586 7-39992930 nan nan
7-39992930 A G 0 0 0 0 3172 0.000239596 7-39992946 nan nan
7-39992946 G C 0 0 0 0 3172 1.57629e-07 7-39992986 nan nan
7-39992986 G A 0 0 0 0 3172 9.91488e-05 7-39993073 nan nan


If I replace the all zeros with a float (e.g. 0.8) I get a result:

Code: Select all

name A1 A2 Freq1 MAF Quality Rsq n Mean_predictor_allele position beta_SNP_add sebeta_SNP_add
7-39992470 A G 0.8 0.8 0.8 0.8 3172 7.69231e-05 7-39992529 -3.89076 3.62736
7-39992529 C T 0.8 0.8 0.8 0.8 3172 9.45776e-07 7-39992552 39.3586 237.176
7-39992552 A T 0.8 0.8 0.8 0.8 3172 0.000214061 7-39992766 -3.34807 4.99724
7-39992766 C T 0.8 0.8 0.8 0.8 3172 0.000245586 7-39992930 1.53096 0.963523
7-39992930 A G 0.8 0.8 0.8 0.8 3172 0.000239596 7-39992946 1.99275 1.95281
7-39992946 G C 0.8 0.8 0.8 0.8 3172 1.57629e-07 7-39992986 1098.86 887.211
7-39992986 G A 0.8 0.8 0.8 0.8 3172 9.91488e-05 7-39993073 -16.8177 10.4208
7-39993073 C G 0.8 0.8 0.8 0.8 3172 0.178307 7-39993085 -0.0320401 0.0307155


The beta, sebeta or other values do not change if I change the value of the float (as expected).

You may want to clarify this in the manual.

Vince

Re: Zeros in mlinfo file results in NAN beta and sebeta

Posted: Fri Jul 13, 2012 3:03 pm
by lckarssen
Thanks for reporting this bug! This helps us making ProbABEL a better product.

I have confirmed your findings and created a bug report at https://r-forge.r-project.org/tracker/i ... &atid=2058 with bug ID #2147.

Re: Zeros in mlinfo file results in NAN beta and sebeta

Posted: Wed Oct 03, 2012 11:04 pm
by lckarssen
Just some quick feedback: this bug is now fixed and will be in the next ProbABEL release (see SVN r.971).

Re: [SOLVED] Zeros in mlinfo file results in NAN beta and sebeta

Posted: Thu Mar 10, 2016 6:06 pm
by ngaddis
FYI - it appears that this bug is back in v0.4.5.

Thanks,
Nathan

Re: [SOLVED] Zeros in mlinfo file results in NAN beta and sebeta

Posted: Thu Mar 10, 2016 8:54 pm
by lckarssen
Thank for reporting this.

The thing is that we use the Rsq column to filter out badly imputed SNPs and simply drop them from the analysis in order to not waste any time on useless computations.

So the bug was actually fixed by updating the documentation (although having a look at it now, I could be more explicit in the text), not be changing ProbABEL's behaviour.

So in our view setting betas and se_betas to 'nan' in case Rsq < 1e-16 is a feature, not a bug. But of course this needs to be mentioned in the documentation.