Animal Model with SNP Based Relatedness NOT pedigree

[Gab Diesel]

Hi breedR users,

I am trying to use breedR and the animal model to estimate the additive genetic variance of a complex oak diseases phenotype however I am struggling to make it work.

We have generated WGS data for over 400 oak trees of species Q. Robur and Q. Petraea across 5 UK woodlands affected by a complex disease and we identified a genome wide set of SNPs. We do not have any pedigree information about the trees however we calculated relatedness using the SNPs markers with the package synbreed.

Now I simply want to used the animal model by replacing the pedigree matrix in the formula with the SNP based relatedness matrix but I am unable to make breedR work and the package throws hard to debug errors after numerous attempts. It is worth noting that my phenotype is categorical, case and control (the trees have simply classified as symptomatic or asymptomatic).

I believe that I need to use the generic model (Page 28 of BreedR Overview PDF) for this but after quite a few tries I am still unable to make it work.

In a few words, is there any guidance about how to use a SNP based relatedness matrix rather than pedigree? Is it possible to use a categorical phenotype like mine in breedR?

Just to give a bit of extra information, I was trying to follow this example for my mode (even though I do not have gg)l:

res.blg <- remlf90(fixed = phe_X ~ gg,

generic = list(block = list(inc.mat,

cov.mat)),

data = globulus)

I tried to input my SNP relatedness matrix as inc.mat, however I am quite unsure on what the package wants as cov.mat. I am also a bit unsure on how to pass the fixed effect my categorical phenotype to the function. I am trying to build a simple model as proof of concept and at the moment I am interested in only having a fixed effect (phenotype) and a ranom effect (the SNP based relatedness).

I know that a recent publication on Oaks (https://www.biorxiv.org/content/10.1101/501387v1.full) has used breedR for this specific task therefore I am trying to reproduce their method.

I can share my code, data and report the errors I get with my attempts if necessary.

Regards,

Gabriele

[Facundo Muñoz]

Dear Gabriele,

First of all, breedR does not handle categorical responses, particularly binary.

For what is worth, if you had a quantitative response, the way to do what you wanted is indeed the generic model, except that the relationship matrix goes to cov.mat. inc.mat is the incidence matrix, relating the observations with the genetic values. In your case I beleive, you have une observation per tree, so 400 observations and a 400x400 relationship matrix. In this case, inc.mat is simply an identity matrix (provided the observations are sorted in the same order as the individuals in the rows/cols of the relationship matrix).

Hope it helps. Sorry it won't work for binary observations.

ƒacu.-

--
Report issues at https://github.com/famuvie/breedR/issues
---
You received this message because you are subscribed to the Google Groups "breedR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to breedr+un...@googlegroups.com.
To post to this group, send email to bre...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/breedr/70f02914-ff3c-4e42-abd0-bb08f31a2011%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Gab Diesel]

Dear Facundo,

Many thanks for the reply. It makes sense now and I understood how to use a marker based matrix in breedR now.

It's a pity that breedR can't handle categorical responses but we are in the process of developing an index for the phenotype under investigation using a variety of descriptors so hopefully we will be able to use breedR in the future for this or other interesting traits. 

Thank you,

Gabriele

To unsubscribe from this group and stop receiving emails from it, send an email to bre...@googlegroups.com.