DO mice, cross info, and "ngen" question

[Mark Sfeir]

I noticed in the qtl2data/Doex/Doex_covar.csv file that the only cross information (as far as I can tell) for the DO mice comes in the form of the number of generations, "ngen", in the breeding program, and no separate cross_info file is present.

Combining this observation with the guideline in the Kbroman.org/qtl2/assets/vignettes/input_files materials about how cross_info information for more complex crosses necessarily "spans multiple columns" and is required to be set up in advance (as no translation of encodings will be performed), I have a question about how DO mouse cross info is handled:

Since DO is a relatively complex cross system among CC mice, is qtl2 specifically designed to use ngen information in relation to the DO mice or other similar crossing programs to simplify specification of cross info?

That ^^ is my main question, but on a more speculative side note, in a DO study, how helpful could it be toward strengthening the QTL analysis to have recorded information on the pedigree and breeding of each and every mouse, starting with the initial CC founders as they interbreed approximately randomly, and continuing over the generations as a given DO stock proliferates? JAX lab may not record this information or make it available for each mouse, and it may be difficult to do so, but I wonder how much more useful it could be in attributing phenotypic effects to a specific QTL variant in a given founder line than keeping track only of the number of generations in the DO breeding program. 

Any guidance on good intro materials to strengthen my theoretical understanding of QTL and what qtl2 does would be appreciated—I plan on reviewing the following intro to QTL mapping w/ model organisms that you gave soon, for instance:

https://biostat.wisc.edu/~kbroman/talks/Louisville/qtllect_ho.pdf

-Mark

[Karl Broman]

Aspects of the DO crossing design are hard-coded into R/qtl2, namely the numbers of generations that the pre-CC lines were at before they entered into the random mating generations. But then we just assume an infinite random mating population, and just need to know the generation a DO mouse is drawn from. When a cross type has just a single column of cross information, it can be included among the covariates rather than having a separate cross info file.

Personally, I don't think that the detailed pedigree will provide much additional value. It can be used either in understanding the genotype/crossing over pattern in the reconstruction of haplotypes, or in the kinship matrix that is used in the polygenic model. But for haplotype reconstruction it doesn't seem like the model needs to be very precise, and in the kinship matrix it seems like the observed genotype sharing, estimated from the data, is more important than the expected sharing, which is all that can be learned from the pedigree.

karl

[Dan Gatti]

I agree with Karl. I don’t think that pedigree information would improve the mapping. We’re performing linkage mapping using the inferred haplotypes from the HMM. As long as those are good, we’re associating phenotypes with founder alleles at each marker.  I’m also not sure that we even have detailed pedigree information. We randomize breeding pairs carefully at each outcrossing generation, and we track siblings when shipping, but otherwise I’m not sure how much detail is retained.

 

Dan

--
You received this message because you are subscribed to the Google Groups "R/qtl2 discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rqtl2-disc+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/rqtl2-disc/593a8c37-7634-4cc6-aae7-f853aa467f40n%40googlegroups.com.

---

The information in this email, including attachments, may be confidential and is intended solely for the addressee(s). If you believe you received this email by mistake, please notify the sender by return email as soon as possible.

[Mark Sfeir]

Thank you both for the information! This definitely adds texture to my understanding of how qtl2 works with DO mice. 

Another question related to the assumption of an infinite random mating population among DO's in qtl2: I've learned that, among CC founders, NZO mice don't tend to produce viable offspring with certain other strains (at least not in an F1 outcross)—has it been found that this confounds any analysis insofar as it makes it less random which mice are featured in the breeding program, or maybe it makes attributing a given QTL effect to NZO variants or variants from other incompatible mice more difficult? Maybe there is a way to manage this? 

Mark

[Dan Gatti]

About 5 to 10% of the DO matings are unproductive. We don’t know why, so it’s hard to ascribe to NZO. In spite of the careful randomized mating, allele frequencies in the DO are drifting form the ideal  1/8^th contribution at all loci. A low founder allele contribution at a given locus makes it harder to detect and effect driven by that founder allele.  I’m not sure what to do about this from a QTL mapping standpoint, other than what we do already. But if this is the direction of your research, you could dive into it and see what you come up with.

 

Dan

To view this discussion on the web visit https://groups.google.com/d/msgid/rqtl2-disc/527803cf-f5b8-407a-9f76-5dd874e1cd97n%40googlegroups.com.

[Mark Sfeir]

That is very interesting! Thanks for this, too, I do think it could be interesting how meiotic recombination mechanisms in relation to elements of different mouse genomes may cause conflict and lead to a trend away from the ideal 1/8th contribution among viable offspring that would hold if randomness were retained between rounds of mating. I don't have great statistical expertise or theoretical foundations here, and as a research tech I am going to be constrained in what I focus on, but it's definitely good to have in mind and seems like something worth exploring.

-Mark