Unusually high LOD scores

[JC]

Hello Group!

We are currently using R/QTL to map grain quality traits in a BC1F2
population, and when we use the permutation function of cim() with
1,000 permutations, we are still seeing LOD scores that feel inflated.
Sometimes as high as 6 or 7. We believe perhaps that the test will
believe it is detecting too many false associations (commiting Type 1
and/or Type 2 errors at a higher rate than expected), because our
allele frequencies are not the same as an F2 and it is compensating by
elevating the empirical LOD threshold. Would you hypothesize this to
be the case? Has anyone seen this before? Is there a way to tell R/QTL
what the expected allele frequencies should be?

Any insights would be much appreciated!

[Karl Broman]

Differences in expected allele frequencies shouldn't have much influence on things unless you have a lot of missing genotype data or are using selective genotyping.

The main thing that I would be concerned about would be relationship structure in the BC1F2 population: batches of individuals with closer relationships among individuals within batches than between batches. This can lead to inflated LOD scores.

karl

> --
> You received this message because you are subscribed to the Google Groups "R/qtl discussion" group.
> To post to this group, send email to rqtl...@googlegroups.com.
> To unsubscribe from this group, send email to rqtl-disc+...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/rqtl-disc?hl=en.
>

[Mac]

With regard to the allele frequency issue, could you code and analyse
the data as coming from a 4-way cross in which three of the parents
happened to be identical? That is as: (AxB)x(AxA).

Ian

[Karl Broman]

(AxB)x(AxA) would be the same as a backcross, though allowing sex differences in the genetic map.

I like your idea, of using the 4-way cross, for handling the allele frequency issue. But I'd do it as the next generation. You just need genotypes on the BC1 parents.

karl