Tutorial on artifacts

[kbroman]

Add a tutorial on identifying artifacts, particularly in multiparent populations.

• single outlier
• missing genotype
• especially narrow LOD peaks
• crazy coefficient estimates
• at a SNP that is monomorphic in the population, but where there is still some variation in the genotype probabilities.

[HannahVMeyer]

Hi - thanks for a great package and providing genotype data for the CC, this has really helped our analyses.

I have come across some very strange effect size estimates and found this open issue. Do you have any advice on how to debug this/what the cause might be?

Below are visualizations of the output from scan1 and scan1coef run on 42 CC strains with 2-4 replicates each of a continuous, but 0-1 bound phenotype (derived from a proportion), loco kinship and batch as covariate. I used the genotype information from https://raw.githubusercontent.com/rqtl/qtl2data/main/CC/cc.zip. For simplicity I show only chr18 (weird effect coefficients) and chr19 (coefficients as might be expected), but this weird pattern is also seen on other chromosomes.

As suggested in this issue, I also tried using the clean_genotypes function. This changed the coefficient estimates (see plot below) but it still doesn't look right:

I am using qtl2_0.32 on R 4.2.1

Many thanks!

[kbroman]

I don't much like scan1coef(). The artifacts are happening in a position without particularly large LOD scores; to me, there's not much reason to be interested in the estimated effects at a position that is not really the QTL.

I would pull out the genotype probabilities at the estimated QTL position and use fit1() to get the estimated effects at that position.

[HannahVMeyer]

Many thanks, Karl! I only showed the plots above for convenience of screenshotting, the actual phenotype has a LOD > 6, significant via permutation testing.

Thanks for pointing me to fit1, I hadn't come across that function in the tutorials I followed. What I liked about the output scan1coef was the ability to plot the estimated coefficients underneath the LOD plot. Is there a similar nice visualisation that comes for fit1?

As for interpreting the fit1 results (and apologies if this is obvious): can I interpret the coef as the effect size per genotype? By then checking the genotypes of the individual strains, I can infer which strain(s) were driving the associations based on genotype?

[kbroman]

I use ciplot() from my broman package.

The estimated effects are for the additive effects coefficients in a linear regression, shifted so that they sum to 0.

[HannahVMeyer]

Thanks very much for the pointers and the speedy reply!