High LOD scores across across the entire linkage map

[Sam Mantel]

Hi Karl,

Thanks so much for maintaining this group!

I am using R/qtl2 to map in a RIX (recombinant inbred intercrossed line) population made from intercrossing individuals from a NAM population with three groups. 

To do this I am using genail4 as the cross type as there are 4 original parental lines (the common maternal parent in the NAM pop and the three paternal parents) and coding genotypes by including four 'markers' at each position (recommended in a previous thread since markers must be biallelic), one that indicates genotype in terms of maternal vs paternal alleles, and the other three indicating which paternal allele is present, if any. This setup gives me the genotype probabilities that I expect and seems to work fine, but wanted to include the info in case its related to the issues I'm having.

For some of the phenotypes I'm mapping when I run a single interval scan with scan1 I get high LOD scores across the entire linkage map with a large proportion of the genome over the significance threshold (see CW16WW_scan1.pdf attached, I'm calculating significance via permutation: red line shows 0.05 cut off, blue line is 0.1, and green line is 0.15). This leads to 21 'significant' peaks at the 0.05 level in this example, which I'm obviously suspicious of.

Any ideas what could be causing this? It does not occur with every phenotype, but does seem to happen with phenotypes that I expect to be very polygenic, like flower size, which has been previously mapped to 15+ small effect QTL in other crosses in this system (CW16WW_scan1.pdf is for corolla width). This makes me feel like at least part of it could be because large portions of the genome are genuinely associated with flower size, but I don't really understand why the significance thresholds would be so comparatively low, and why no part of the genome would have a LOD below ~4. 

I have tried including a covariate to the model to correct for what I'm calling "Cross Group", which is a categorical variable indicating which NAM groups were crossed to produce any given RIX individual (CW16WW_scan1_CG.pdf), and have also tried running an LMM including a kinship matrix (CW16WW_scan1_LOCO.pdf) both of which partially fix the problem, and give more reasonable looking results, but also give different results.

Any insight or recommendations you could give would be much appreciated.

Thanks so much,

Sam

[Sam Mantel]

Looks like one of the plots didn't get attached - here it is!

[Karl Broman]

I think the genotypes are confounded with cross, and so I would include cross covariate indicators in scan1().

karl