Structure with Phenotye data and ln P(D) problems

[Victoria Springer]

Hello. I am a biological anthropologist and would like to use
Structure with some discrete presence/absence morphology traits. I
have run the analysis many different ways and no matter what I do, I
get similar results with the Ln Like in the burnin and the Est Ln P(D)
in the iterations. The weird results show up in the bottom running
scroll and in the data plots of the likelihood and Ln P(D). Every
time, the Ln Like in the burnin is constant 2.0000000000000013 x
10^264. Then at the end of the burnin, the Est Ln P(D) is always 0.
So the likelihood plot is a straight horizontal line to the end of the
burnin, a straight vertical line to 0, then a straight horizontal line
to the end of the iterations. The Ln P(D) plot is blank.

For this particular run, I got these summary results:
Estimated Ln Prob of Data = -898.8
Mean value of ln likelihood = -851.8
Variance of ln likelihood = 94.1
Mean value of alpha = 0.2388

Mean value of Fst_1 = 0.3542
Mean value of Fst_2 = 0.1273

And the bar plot looks like what I would expect knowing my populations
and data.
Anybody have any suggestions or ideas of what this means or what to do
about it? Thanks.
-Victoria

[Vikram Chhatre]

Hi Victoria,

I am not sure why you are getting such a result. But if you posted
the plots, and provided some more information on your run, someone may
be able to help.

You can post the run parameters and runtime output on external website
like http://www.pastebin.com for others to view.

V

> --
> You received this message because you are subscribed to the Google Groups "structure-software" group.
> To post to this group, send email to structure...@googlegroups.com.
> To unsubscribe from this group, send email to structure-softw...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/structure-software?hl=en.
>
>

[Victoria Springer]

Thanks. Let's see if I can paste something in here. I got the
runtime output to paste, but I can't figure out how to get a figure to
paste here or in pastebin.

Here is some runtime output from a short run - the basics are similar
to what happens with longer runs.

Finished initialization; starting MCMC
500 iterations + 500 burnin

Rep#: Alpha F1 F2 Ln Like
100: 0.546 0.392 0.206
7147212327398288900000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
200: 0.456 0.184 0.331 71472123273982889(same as above)
300: 0.649 0.176 0.596 71472123273982889(same as
above)
400: 0.331 0.282 0.504 71472123273982889(same as above)
500: 0.339 0.257 0.433 71472123273982889(same as above)

BURNIN completed

Rep#: Alpha F1 F2 Ln Like Est Ln P(D)
600: 0.239 0.228 0.308 -840 0
700: 0.241 0.242 0.302 -856 0
800: 0.154 0.221 0.099 -855 0
900: 0.086 0.398 0.194 -842 0
1000: 0.177 0.302 0.357 -846 0

MCMC completed

Run parameters:
127 individuals
14 loci
2 populations assumed
500 Burn-in period
500 Reps

--------------------------------------------
Estimated Ln Prob of Data = -881.8
Mean value of ln likelihood = -849.1
Variance of ln likelihood = 65.5
Mean value of alpha = 0.1991

Mean value of Fst_1 = 0.2373
Mean value of Fst_2 = 0.2981

--------------------------------------------
Proportion of membership of each pre-defined
population in each of the 2 clusters

Given Inferred Clusters Number of
Pop 1 2 Individuals

1: 0.527 0.473 29
2: 0.171 0.829 38
3: 0.817 0.183 60

Allele-freq. divergence among pops (Net nucleotide distance),
computed using point estimates of P.

1 2
1 - 0.1555
2 0.1555 -

Average distances (expected heterozygosity) between individuals in
same cluster:
cluster 1 : 0.2974
cluster 2 : 0.2933

On Nov 18, 9:39 pm, Vikram Chhatre <crypticline...@gmail.com> wrote:
> Hi Victoria,
>
> I am not sure why you are getting such a result.  But if you posted
> the plots, and provided some more information on your run, someone may
> be able to help.
>
> You can post the run parameters and runtime output on external website

> likehttp://www.pastebin.comfor others to view.

[Emily]

Hi! I have the same problem as Victoria. I am using the thrush example data set (in teaching). I am wondering if the problem has anything to do with the installation. I recently had v. 2.3.3. added to machines in our teaching classroom and I do not have full administrator rights. I had problems first getting the program to work. Now things seem to work, but the output is suspicious: Like Victoria I have at the end of the burnin, Est Ln P(D) is always 0.
and the likelihood plot is a straight horizontal line to the end of the burnin, a straight vertical line to 0, then a straight horizontal line to the end of the iterations.  The Ln P(D) plot is blank. See the pics above. Nevertheless there is some summary data for LnP(D):
Estimated Ln Prob of Data   = -2753.0
Mean value of ln likelihood = -2716.6
Variance of ln likelihood   = 72.8
Mean value of alpha         = 0.0454

Mean value of Fst_1         = 0.1809
Mean value of Fst_2         = 0.0613

BUT, I don't know if I can trust it. Any suggestions?

 

[Emily]

Hi again,

I have just run the same files in structure 2.2. Now, I have the Est LnP(D) reported for each rep (and nice graphs). The good news is that the summary reported in 2.3.3 is pretty much the same as I got for the runs done in structure 2.3.3 so, maybe there is just some glitch in the recording of those values per replicate of the mcmc. It would be nice to have at least the graph feature working in 2.3.3 so you could evaluate how well you have estimated the burn in.So, I hope Pritchard et al are reading these posts now and then.

Emily

[Victoria Springer]

This is really interesting. Thanks for this info. It's good to hear
that it probably isn't an issue with my data, since I'm also using
version 2.3.3. I'll have to see if I can come up with a different
version.
-Victoria

[Victoria Springer]

I'm reviving this thread to see if anyone has any suggestions on where
I could locate an earlier version of Structure to test it out with my
data as Emily was able to do in this thread. Anyone?

On Nov 18, 9:39 pm, Vikram Chhatre <crypticline...@gmail.com> wrote:

> Hi Victoria,
>
> I am not sure why you are getting such a result.  But if you posted
> the plots, and provided some more information on your run, someone may
> be able to help.
>
> You can post the run parameters and runtime output on external website

> likehttp://www.pastebin.comfor others to view.

[Vikram Chhatre]

Victoria -

Earlier versions of Structure available on this page:

http://pritch.bsd.uchicago.edu/structure_software/release_versions/v2.3.3/html/structure.html

Scroll down and look for "Earlier Versions".

V

[Victoria Springer]

Awesome! Thanks.

On Dec 13, 4:19 pm, Vikram Chhatre <crypticline...@gmail.com> wrote:
> Victoria -
>
> Earlier versions of Structure available on this page:
>

> http://pritch.bsd.uchicago.edu/structure_software/release_versions/v2...

>
> Scroll down and look for "Earlier Versions".
>
> V
>
>
>
>
>
>
>
> On Tue, Dec 13, 2011 at 4:17 PM, Victoria Springer <spring...@gmail.com> wrote:
> > I'm reviving this thread to see if anyone has any suggestions on where
> > I could locate an earlier version of Structure to test it out with my
> > data as Emily was able to do in this thread.  Anyone?
>
> > On Nov 18, 9:39 pm, Vikram Chhatre <crypticline...@gmail.com> wrote:
> >> Hi Victoria,
>
> >> I am not sure why you are getting such a result.  But if you posted
> >> the plots, and provided some more information on your run, someone may
> >> be able to help.
>
> >> You can post the run parameters and runtime output on external website

> >> likehttp://www.pastebin.comforothers to view.