summary scantwo/epistasis/LODS

[Stuart]

Hi

I need help to interpret summary of plot.scantwo. I’m interested in
epistatic interactions LOD scores to put in my write up. We did not
choose any LOD scores threshold so we just did it based on 0.05.

Normally for QTL I use LOD score of 2 for significance but if want to
say we have significant epistatic interaction which of these LOD score
I should I use? Fo all of these4 traits I can see significant
interaction between marker in effect plots but I need LOD scores to
include in my write up to make it more statistical meaningful.

Phenotypes chr1 chr2 pos1f pos2f lod.full lod.fv1 lod.int pos1a pos2a
lod.add lod.av1
XX1 4 9 50 80 3.515 1.313 0.345 50 80 3.17 0.968
XX2 3 7 80 60 6.872 2.914 1.090 40 60 5.78 1.824
XX3 4 9 120 20 5.322 1.079 0.027 120 20 5.3 1.053

[Karl Broman]

There is a relatively complete discussion of the interpretation of scantwo results at the following:

http://www.rqtl.org/tutorials/new_summary_scantwo.pdf

In general, I would recommend the use of a fixed threshold (like 2) for LOD scores; the appropriate threshold depends on the nature of the cross, the size of the genome, and the type of LOD score. So while a threshold of 3.5 might be appropriate for a single-QTL genome scan in an intercross, the appropriate threshold for a LOD score assessing an epistatic interaction would be completely different.

I like to use permutation tests to define relevant thresholds (or, even better, to assign genome-scan-adjusted p-values to LOD scores). For these scantwo analyses, that means having to do a permutation test with a two-dimensional, two-QTL genome scan. This can be quite computationally intensive, though if Haley-Knott regression (method="hk" in scantwo) is appropriate for your data, the permutation test would be considerably faster than with other methods.

For the three pairs of loci you displayed, the interaction LOD scores are 0.3, 1.1 and 0.03. None of these are likely to be seen to be significant, and so I am surprised to hear that the interactions are clear in the effect plots.

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.
>

[Karl Broman]

I wrote "In general, I would recommend the use of a fixed threshold
(like 2)..." but I hope it's clear that I meant to say, "In general, I
would recommend *against* the use of a fixed threshold (like 2)..."

I don't want to think about how many times I've made such a mistake
and not caught it.

Karl

[Stuart Broad]

Thanks very much indeed. I can understand what you mean. I am just attching you effect plot and the statistical summary from scan two for the the markers where I can see significant interactiona and very Low LOD score in case of int.lod.

 

Please have look and if possible some feedback.

 

Thanks

[Karl Broman]

The interaction in the effect plot does not look too strong to me. While the lines are not parallel, the standard errors are large, and so these results are not incompatible with additivity.

karl

> <effectplot-summaryscan2.ppt>