structure harvester

[Gabriel Eduardo Melim Ferreira]

Hi,
I have been made many big runs using default settings of front end of
STRUCTURE. I was very comfortable with Structure Harvester (SH) but it
started to not return me the delta K value and advertises that I have
to include more replicates. When it doesn’t work the message is that
SH needs more than 3 replicates to calculate delta K but I uploaded 10
replicates for each K value.
There’s someone else having this problem? Actually it would be great
if someone knows how to deal with this.
Gabriel
Leishmania/ 14 loci/ 180 individuals
Admixture model/ default settings
Burning period 100,000/ running period 1 million
K from 1 to 10/ 10 runs for each K

[Dent]

Apologies if this shows up twice, google groups is acting strangely this morning.

Hey Gabriel,

I'm the author of Structure Harvester (http://taylor0.biology.ucla.edu/structureHarvester/index.php). It sounds like there is either a bug in my code or a problem with the input format. If you send me the archive (.zip) that's giving you trouble off-list, I'll try to diagnose and correct the problem. 

Thanks,

d

[Dent]

We solved this problem off-list. This is a follow up for the rest of the list.

There are tests performed by Structure Harvester to ensure that your data conform to certain requirements before the Evanno method is applied. One of these tests checks to see if for all K, the standard deviation of the ln prob(data) values is not zero since we need to divide by this value. It turns out that this test was too strict, the first K and last K may have standard deviations of zero without causing a divide by zero error (see the faq for details on the delta K calculation). I've made this change to the code.

Thanks Gabriel!

d

[Gabriel Eduardo Melim Ferreira]

Hi Dr Dent,
I’m sincerely gratefully for your help. I also need to thank you for
the SH that’s very useful.
Gabriel

On 6 jul, 14:45, Dent <dent.e...@gmail.com> wrote:
> We solved this problem off-list. This is a follow up for the rest of the
> list.
>
> There are tests performed by Structure Harvester to ensure that your data
> conform to certain requirements before the Evanno method is applied. One of
> these tests checks to see if for all K, the standard deviation of the ln
> prob(data) values is not zero since we need to divide by this value. It
> turns out that this test was too strict, the first K and last K may have
> standard deviations of zero without causing a divide by zero error (see the

> faq <http://taylor0.biology.ucla.edu/~dearl/structDev/faq.html> for details