using snps in known gene as covariate

[elsade...@gmail.com]

I've used qtl2 to perform QTL analysis on data from an 8-parent MAGIC population using crosstype = "risib8"

One genomic region that is significantly associated with a trait is in LD with a gene known to be involved in that trait, though the peak is somewhat distant from the gene. I'd like to determine whether this gene is enough to explain the association, or if there may be more genes involved. I thought performing the genome scan again using the allele at a locus close to the gene as a covariate would accomplish this, but it didn't work how I expected; the association is now stronger. Also, the LOD of the locus I used as the covariate was not 0, as I would have expected.

This makes me thing that either I am not understanding what covariates do, or I've made a coding error. Basically I'm adding a fixed effect, right?

Thanks for the help!

Elsa

[Karl Broman]

It's hard to say what might be going wrong without seeing your code and results.

In treating a position's genotype as a covariate, you need to be sure to include 7 covariate columns. 

And I've found that using the genotype probabilities at a strong-effect QTL may not account for all of the effect. The Haley-Knott regression approximation doesn't work well when the effect is large and you're considering closely linked positions. However, if you do a single random imputation and use those imputed genotypes for the covariate, that can work.

karl

--
You received this message because you are subscribed to the Google Groups "R/qtl2 discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rqtl2-disc+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/rqtl2-disc/9787900a-f4c9-4781-b065-bdbdf1cec023n%40googlegroups.com.

[elsade...@gmail.com]

Thanks for the reply. Why 7 covariate columns?That's probably what I was missing; I used 1 column with a 0 for one allele at the locus and a 1 for the other allele. Because I chose one of the markers which is known for all individuals, I didn't use genotype probabilities, I used the known genotype.

Elsa

[Karl Broman]

Since it’s an 8-way MAGIC population, there are 8 possible alleles at the locus. You include indicators for 7 of the 8 alleles, as the model already includes an intercept.

karl