Interpreting structure harvester results

[Ruud van den heuvel]

Hello,

I've run 6 micro satellites for the fish species Prochilodus lineatus which I've gathered from 3 different locations. I’m running structure with admixture, correlated frequencies, 10 runs for each K, burnin=1000000 and 1000000 steps after that.

An AMOVA tests shows that the most molecular variance can be found within populations. When we look at the results from harvester I'm not sure how to interpret these results because the deltaK for K=3 is very high. However I think that there is probably little to no population structure to be found.

Here is the harvester link:

http://taylor0.biology.ucla.edu/structureHarvester/completedJobs/imperfect-water-10ce/summary.html

Any input would be appreciated.

[Andrea Schreier]

Hi, Ruud! Don't forget that K=1 cannot be included in delta K
calculations so if that is the most likely K, it wouldn't be reflected
in those results. I actually don't see K=1 in your likelihood plot,
did you test K=1 in your analysis?

Good luck!

Andrea

> --
> You received this message because you are subscribed to the Google Groups "structure-software" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to structure-softw...@googlegroups.com.
> To post to this group, send email to structure...@googlegroups.com.
> Visit this group at http://groups.google.com/group/structure-software.
> For more options, visit https://groups.google.com/d/optout.

[Vikram Chhatre]

Yes, you need to test K=1 because Delta K, by the inherent nature of the method, will not show estimates for the first and last K you test.  Thus, always test K1 through Kn+1.  In other words, if you want to test 8 models, test K1 through K9.

V

[Ruud van den heuvel]

Op vrijdag 18 juli 2014 18:01:31 UTC+2 schreef Vikram Chhatre:

Yes, you need to test K=1 because Delta K, by the inherent nature of the method, will not show estimates for the first and last K you test.  Thus, always test K1 through Kn+1.  In other words, if you want to test 8 models, test K1 through K9.

V

Hello Vikram and Andrea

I've also included K=1 into the structure analysis. 

http://taylor0.biology.ucla.edu/structureHarvester/completedJobs/snowy-firefly-7591/summary.html

am I correct to assume that K=3 is the most likely amount of populations within my testgroup based on the deltaK?

 

On Fri, Jul 18, 2014 at 11:24 AM, Andrea Schreier <amdr...@gmail.com> wrote:

Hi, Ruud!  Don't forget that K=1 cannot be included in delta K
calculations so if that is the most likely K, it wouldn't be reflected
in those results.  I actually don't see K=1 in your likelihood plot,
did you test K=1 in your analysis?

Good luck!

Andrea

On Fri, Jul 18, 2014 at 2:31 AM, Ruud van den heuvel
<ruudvd...@gmail.com> wrote:
> Hello,
>
> I've run 6 micro satellites for the fish species Prochilodus lineatus which I've gathered from 3 different locations. I’m running structure with admixture, correlated frequencies, 10 runs for each K, burnin=1000000 and 1000000 steps after that.
>
> An AMOVA tests shows that the most molecular variance can be found within populations. When we look at the results from harvester I'm not sure how to interpret these results because the deltaK for K=3 is very high. However I think that there is probably little to no population structure to be found.
>
> Here is the harvester link:
>
> http://taylor0.biology.ucla.edu/structureHarvester/completedJobs/imperfect-water-10ce/summary.html
>
> Any input would be appreciated.
>
> --
> You received this message because you are subscribed to the Google Groups "structure-software" group.

> To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsub...@googlegroups.com.

> To post to this group, send email to structure...@googlegroups.com.
> Visit this group at http://groups.google.com/group/structure-software.
> For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "structure-software" group.

To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsub...@googlegroups.com.

[Vikram Chhatre]

K=3 doesn't agree with lnPr(K).  Let's look at the bar plots for K=2 and K=3.

V

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

[Vikram Chhatre]

As you can see, there is no discernible pattern of dominant cluster memberships in any individuals either at K=2 or K=3.  Your data does not seem to have structure, or has weak structure.  If this is a very large dataset, you can try to partition and try testing individual parts for weak structure.

On Mon, Jul 21, 2014 at 10:00 AM, Ruud van den heuvel <ruudvd...@gmail.com> wrote:

I took the bar plots for K2 and K3 for three different runs.

Op maandag 21 juli 2014 15:03:10 UTC+2 schreef Vikram Chhatre:

> To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsubscribe@googlegroups.com.

> To post to this group, send email to structure...@googlegroups.com.

> Visit this group at http://groups.google.com/group/structure-software.
> For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "structure-software" group.

To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsubscribe@googlegroups.com.

To post to this group, send email to structure...@googlegroups.com.

Visit this group at http://groups.google.com/group/structure-software.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "structure-software" group.
To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsub...@googlegroups.com.
To post to this group, send email to structure...@googlegroups.com.
Visit this group at http://groups.google.com/group/structure-software.
For more options, visit https://groups.google.com/d/optout.

[Vikram Chhatre]

To clarify the delta K result, note the extent to which Delta K is varying (y axis).  When real structure is present, the scale is much larger (here only 0-20).

V

[Ruud van den heuvel]

The amount of individuals is only 201 so this makes for a small data set. I can try running structure in LOCPRIOR mode because I have the coordinates for the sample sites. Do you reckon that this will make any difference?

Greetings,

Ruud

Op maandag 21 juli 2014 16:13:30 UTC+2 schreef Vikram Chhatre:

[Vikram Chhatre]

I doubt it, but give it a shot.

V

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

[Ruud van den heuvel]

These are the results when geographical information is included

http://taylor0.biology.ucla.edu/structureHarvester/completedJobs/wispy-plateau-be3e/summary.html

There still seems to be no structure in my data.

Thank you for all the help

Op maandag 21 juli 2014 16:42:03 UTC+2 schreef Vikram Chhatre:

[Nikki]

Hey Vikram,

Two questions: 1) concerning your comment about the y-axis of Delta K, I completely agree that when the scale is "low" (e.g. values <10) that that is evident of little/no real structure. Do you know of any data/publications to reference that support that? And 2) can the same be said for when the scale is low for mean LnP(K) (I am calling "low" to be a scale in the -1000's as opposed to the -10000s)?

Thanks!

Nikki