Haplotype-based QTL vs SNP-based

[Billy Jiang]

Dear Karl, 

I trust this finds you well. 

I've been trying to do QTL mapping in diversity outbred mice on a few phenotypes, let's say fasting insulin as an example. I performed haplotype-based (using scan1) and SNP-based (using scan1snps) QTL mapping across the entire genome. However, I found some cases where there is a huge peak QTL (LOD = 12) in haplotype-based mapping, but when I looked at the SNP-based result, there was no peak at all (LOD < 3). There are also cases where there was a high peak at SNP-based mapping (LOD > 7), but no haplotype-based peak (under the permutation threshold) at all. Do you know why this was the case? And what might've caused this inconsistency? 

Thank you so much!

Cheers,

Billy

[Karl Broman]

The haplotype-based LOD scores should each be larger than SNP-based LOD scores, as the haplotype-based ones involve a contrast across the 8 haplotypes, while the SNPs will partition those haplotypes into two groups (the founders with one SNP allele and the founder with the other SNP allele).

The haplotype-based LOD scores can be considerable larger than any single SNP, if the pattern of phenotype effect is not well explained by a single SNP.

If analysis of individual SNPs are really giving strong association signals that aren't present in the haplotype-based results, that could be because of the smoothing involved in the haplotype inference. Another possibility is that the haplotype analysis is using an additive allele model (with 7 degrees of freedom) while the SNP analysis is allowing dominance (2 degrees of freedom tests).

If you look at Figure 1 in the original R/qtl2 paper (https://doi.org/10.1534/genetics.118.301595), you'll see the pattern, that the LOD scores are larger for the haplotype-based analyses (Fig 1B), but that single SNP results may show larger peaks in somewhat different places (Fig 1C).

   https://academic.oup.com/view-large/figure/398119075/495f1.jpeg

karl

[Billy Jiang]

Thank you!