[adegenet-forum] seeking advice on how to use DAPC grouping in association genetics

[Matthieu BOGARD]

Dear Thibaut,

Thank you very much for your answer. I hope you don't mind if I ask you
some precisions:

At the moment, I used the following code to extract individuals
coordinates but I'm not sure this is what you meant when you said "As an
alternative, you can use the discriminant functions of the DAPC":

gen <- df2genind(gen, sep="", ploidy=2)
grp <- find.clusters(gen)

dapc2 <- dapc(gen, grp$grp, n.da=100, n.pca=50)
temp <- optim.a.score(dapc2)
OptimalNbPCA <- temp$best

dapc3 <- dapc(gen, grp$grp, n.da=100, n.pca=OptimalNbPCA)
dapc3$ind.coord

If this is wrong, could you please precise how to extract the
discriminant functions of the DAPC?

Cheers,
Matthieu


Le 23/11/2013 17:07, Jombart, Thibaut a écrit :
> Hi there,
>
> there are several ways you can do this.
>
> First, the groups being known, you can simply remove entirely all potential stratification by regressing your genetic data (or your response variable) onto the group membership vectors using a simple ANOVA (and keeping the residuals of the corresponding model). This could be an overkill though, as stratification may be mainly concerning sets of your groups (expl: groups 1-2 vs groups 3-5). As an alternative, you can use the discriminant functions of the DAPC as regressors, and keep the residuals.
>
> In any case, the operation will look like:
>
> new.x <- residuals(lm(x ~ myRegressor))
>
> Please consider using the forum for such questions - other people may be interested in the answer. See section 'contacts' on:
> http://adegenet.r-forge.r-project.org/
>
> Cheers
> Thibaut
>
>
> ________________________________________
> From: Matthieu BOGARD [matthieu.bogard at clermont.inra.fr]
> Sent: 21 November 2013 16:04
> To: Jombart, Thibaut
> Subject: seeking advice on how to use DAPC grouping in association genetics
>
> Dear Thibault,
>
> I'm currently carrying an association genetics study on earliness in
> wheat. I've been looking at stratification of my panel with the DAPC
> method you described in Jombart et al. 2010.
>
> Could you please advise me on the best way to take into account panel
> stratification in the model testing single marker association?
>
> At the moment, I was thinking of using the group membership
> probabilities output in DAPC{} of the k-1 groups (with k the total
> number of groups) as people usually do with group membership
> probabilities output in STRUCTURE (Price, Pritchard et al. software). Do
> you think that would be right or would you prevent doing this? Is there
> another way to do so by using the PCA coordinates of individuals for
> example?
>
> Your help would be much apreciated.
>
> Kind regards,
> Matthieu BOGARD
>
>