[adegenet-forum] adegenet-forum Digest, Vol 131, Issue 2

[Zhian Kamvar]

1. xvalDAPC() require one of the input as group membership (grp) of
individuals and then it gives number of optimum PC as one of the output.
However the grp what we used is from find.clusters(). Do I need to use
optimum PC from xvalDAPC and re-run find.clusters()?


No. With find.clusters(), you want to use as many principal components as
you can for the k-means clustering. The purpose of xvalDAPC() is to
determine the optimum number of PC to perform the DAPC, not any other
analysis. This is to prevent over-fitting the data.

2. How can I use posterior probabilities from DAPC analysis? The prior and
posterior group membership changing for few individuals

There's a lot you can do with the results from the DAPC. You can use
summary() on your result to give you an overall or by cluster reassignment
rate. You can also use the loadings from the discriminant axes to
investigate what alleles are influencing the separation. I would recommend
going through the tutorial:
https://github.com/thibautjombart/adegenet/blob/master/tutorials/tutorial-dapc.pdf

If you want an indicator of how well the clustering separated the sample by
genetic distance, you can use AMOVA and report the phi statistic (but not
the p-value, see: https://doi.org/10.1111/mec.13243)

3. How can I select optimum number of DA functions? I used default
(n.da=grp-1) in DAPC analysis. How to decide how many DA functions is
required? Is there any statistic to take account of

I believe this is covered in the DAPC tutorial, but in short, the barplot
represents the amount of separation between groups along the axes of the
DAPC; choose the largest ones. The first one always represents the most
separation, the second one, the second most, and so on. For example, If you
see  three large bars and several smaller trailing bars, then that
indicates that most of the variation is described by the first three axes
and you only need those three in the model (the rest will only provide
noise). When you run a scatterplot from DAPC, it will always show you the
first two discriminant axes.

Hope that helps.

Best,
Zhian

On Thu, Oct 24, 2019 at 6:26 PM Das, Roma (ICRISAT-IN) <r.das at cgiar.org>
wrote:

> Thanks Zhian for your reply. Really helpful. Please help me to further
> understand this
>
>
>
> 1. xvalDAPC() require one of the input as group membership (grp) of
> individuals and then it gives number of optimum PC as one of the output.
> However the grp what we used is from find.clusters(). Do I need to use
> optimum PC from xvalDAPC and re-run find.clusters()?
>
>
> 2. How can I use posterior probabilities from DAPC analysis? The prior and
> posterior group membership changing for few individuals
>
>
>
> 3. How can I select optimum number of DA functions? I used default
> (n.da=grp-1) in DAPC analysis. How to decide how many DA functions is
> required? Is there any statistic to take account of
>
>
>
>
>
> Thanks and regards,
>
> Roma
>
>
>
> *From:* adegenet-forum [mailto:
> adegenet-forum-bounces at lists.r-forge.r-project.org] *On Behalf Of *Zhian
> Kamvar
> *Sent:* 24 October 2019 10:15
> *To:* adegenet-forum at lists.r-forge.r-project.org
> *Subject:* Re: [adegenet-forum] adegenet-forum Digest, Vol 131, Issue 2
>
>
>
> Hello Roma,
>
>
>
> Use the groups from find.clusters.
>
>
> It's a common misconception, but DAPC is not a method to define groups. It
> is a tool that allows you to create a model of your data based on your
> groups so that you can assess how well you can differentiate samples into
> individual groups (similar to AMOVA) and give you a method to predict what
> groups your samples belong in based on that model.
>
> find.clusters() and snapclust() are the only functions in adegenet that
> can determine groups de novo from your data.
>
> Hope that helps,
>
> Zhian
>
>
>
>
>
> On Thu, Oct 24, 2019 at 11:00 AM <
> adegenet-forum-request at lists.r-forge.r-project.org> wrote:
>
> Send adegenet-forum mailing list submissions to
>         adegenet-forum at lists.r-forge.r-project.org
>
> To subscribe or unsubscribe via the World Wide Web, visit
>
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/adegenet-forum
>
> or, via email, send a message with subject or body 'help' to
>         adegenet-forum-request at lists.r-forge.r-project.org
>
> You can reach the person managing the list at
>         adegenet-forum-owner at lists.r-forge.r-project.org
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of adegenet-forum digest..."
>
>
> Today's Topics:
>
>    1. DAPC-Find optimum number of groups (Das, Roma (ICRISAT-IN))
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Thu, 24 Oct 2019 07:59:10 +0000
> From: "Das, Roma (ICRISAT-IN)" <r.das at cgiar.org>
> To: "adegenet-forum at lists.r-forge.r-project.org"
>         <adegenet-forum at lists.r-forge.r-project.org>
> Subject: [adegenet-forum] DAPC-Find optimum number of groups
> Message-ID:
>         <
> DB6P195MB04211B823F362C731D17B659FA6A0 at DB6P195MB0421.EURP195.PROD.OUTLOOK.COM
> >
>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Hello everyone,
>
> *         Based on DAPC analysis, I am not sure whether I should treat the
> final group for individuals line as 1) prior group from find.clusters() or
>
>        2) group with maximum posterior probability after xval.DAPC()
>
>
> As in scatterplot from DAPC analysis  individuals are plotted based on
> prior group. Please advise if there a way to choose optimum number of
> discriminating functions to be used.
>
>
>
>
> Regards,
>
> Roma
>
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <
> http://lists.r-forge.r-project.org/pipermail/adegenet-forum/attachments/20191024/5cf7da0c/attachment-0001.html
> >
>
> ------------------------------
>
> Subject: Digest Footer
>
> _______________________________________________
> adegenet-forum mailing list
> adegenet-forum at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/adegenet-forum
>
> ------------------------------
>
> End of adegenet-forum Digest, Vol 131, Issue 2
> **********************************************
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/adegenet-forum/attachments/20191025/a65c954a/attachment.html>