Rqtl

[cyan...@wisc.edu]

Hi, I recently encountered some strange issues with QTL mapping. I have a dataset with lots of 0 and few data points that are above 0. I first tried using the default model="normal", and I got several QTLs and everything seems to work fine. When I started adding more QTL via addqtl, I found another significant QTL that is only 0.7cM away from a QTL found at the beginning. The 1.5 LOD intervals of these two QTL overlap as well. What are the possible reasons for seeing such QTLs?

Also, I looked up QTL mapping for non-normal traits, and I first tried it with model="np". However, I stumbled in the fitqtl part as it only takes in model="normal" or "binary". Should I just try converting my data into 0s and 1s, and do everything with model="binary" instead? Do I lose power for doing it the binary way?


Thanks,
CJ

[Karl Broman]

Artifacts of tightly-linked QTL are a constant problem; I don't have a good solution other than to not trust the case of QTL with <2 typed markers between them, and to delete them "manually".

For phenotypes with many 0's, I generally prefer to use an approach similar to that described in Broman (2003) Genetics 163:1169-1175, http://www.genetics.org/content/163/3/1169.long

Define two derived phenotypes: a binary trait the indicates ==0 or >0, and a quantitative trait where all of the individuals with 0 phenotype are treated as missing values.

karl

> --
> You received this message because you are subscribed to the Google Groups "R/qtl discussion" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to rqtl-disc+...@googlegroups.com.
> To post to this group, send email to rqtl...@googlegroups.com.
> Visit this group at http://groups.google.com/group/rqtl-disc.
> For more options, visit https://groups.google.com/d/optout.

[cyan...@wisc.edu]

Hi Karl,

Thanks for the advice and link! Is it still possible to proceed and do addqtl/fitqtl with the nonparametric/binary/2part method? I encountered some issues when I did the binary one.

Thanks,

CJ

[Karl Broman]

The multiple QTL analysis with fitqtl/addqtl should work with the binary model, but 2part and nonparametric are not implemented.

There's not really a multiple-QTL analogue for the single-QTL nonparametric analysis. But sometimes I convert phenotypes to normal quantiles and then apply the usual "normal" model. R/qtl includes a function nqrank() for this transformation. It's not ideal for the case of many ties at 0.

What issues did you encounter?

karl

[cyan...@wisc.edu]

This was done with binary model, and I got the following error message:

> tips.addqtl <- addqtl(tips, qtl=tips.qtl, formula=y~Q1, model = "binary", method="hk")
There were 50 or more warnings (use warnings() to see the first 50)
> warnings()
Warning messages:
1: In addqtl(tips, qtl = tips.qtl, formula = y ~ Q1, model = "binary",  ... :
  Dropping 13 individuals with missing phenotypes.
2: In fitqtlengine(pheno = pheno, qtl = qtl.obj, covar = covar,  ... :
  Didn't converge.
3: In fitqtlengine(pheno = pheno, qtl = qtl.obj, covar = covar,  ... :
  Didn't converge.
4: In fitqtlengine(pheno = pheno, qtl = qtl.obj, covar = covar,  ... :
  Didn't converge.
5: In fitqtlengine(pheno = pheno, qtl = qtl.obj, covar = covar,  ... :
  Didn't converge.

[Karl Broman]

In fitting a model with a binary trait, the analysis method involves an iterative algorithm.

In most cases, you can ignore these "didn't converge" warnings. They often happen in regions with little evidence for a QTL.

Or you could increase the maxit argument (for the maximum number of iterations), or increase the tol argument (for the tolerance for convergence).

karl

[cyan...@wisc.edu]

Thanks for the tips! I will keep them in mind!

[Elise S]

Hi Karl, 

I also encountered these 'didn't converge' warnings, using 'hk' and binary model in the scanone function.

You indicated in a previous response to this íssue that such warnings 'happen in regions with little evidence for a QTL'. 

Can I somehow check in my data, for which pair of marker this convergence warnings are associated ? 

Is there a way I can inspect/ trim my data to avoid such warnings ?

Thank you in advance for your advice,

Elise