Input data file and run settings

[Ana Mondon]

Hello everyone. I'm starting to use STRUCTURE and have a few basic questions. My data set has 4 sampling locations, and individuals "split" into the same biotypes in each location. How do I specify this to the program? Should I use 12 different integers in the Pop ID column? or just let all individuals to be mixed?

Besides that, what's your suggestion about run times and iterations. I was planning to follow Evanno's (2005) method to select the best K, using 20 iterations for each K, 10000 burnin, 10000 MCMC after burn in. Do I need to do something else after that? Or is it just about running the simulation again for the fixed K?

Thanks a lot!

[Vikram Chhatre]

You can't do a nested analysis.  But if you find that structure is weak with 4 pops, you can always subset the data set and test further for that.

You may be able to get away with 10K burnin, but look at alpha progress to make sure it has converged prior to collecting samples.  I would collect data (mcmc) for at least 50K and preferably 100K. 

If you plan to use deltaK method, doing more Ks than you expect is useful, since the method will omit first and last K tested.

V

--
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.

[Ana Mondon]

Thanks for your answer! As I can't do nested analysis what I did was assigning each of the 12 subsets a number of population (do you think that might hepl?)

For deltaK is selected K=1 to 15 (my number of subsets + 3, as suggested by Evanno et al), Burn-in= 10000 and 10000 MCMC, 20 iterations for each K. As I said, I chose those parameters based on Evanno et al and their method. Not sure if I should run more burn-in processes and MCMC, which I plan to do once I have selected K.

Is there anything I migth be missing? Steps to follow? 

Again, thanks a lot!

El martes, 19 de mayo de 2015, 16:41:11 (UTC-3), Vikram Chhatre escribió:

You can't do a nested analysis.  But if you find that structure is weak with 4 pops, you can always subset the data set and test further for that.

You may be able to get away with 10K burnin, but look at alpha progress to make sure it has converged prior to collecting samples.  I would collect data (mcmc) for at least 50K and preferably 100K. 

If you plan to use deltaK method, doing more Ks than you expect is useful, since the method will omit first and last K tested.

V

On Tue, May 19, 2015 at 3:02 PM, Ana Mondon <anam...@gmail.com> wrote:

Hello everyone. I'm starting to use STRUCTURE and have a few basic questions. My data set has 4 sampling locations, and individuals "split" into the same biotypes in each location. How do I specify this to the program? Should I use 12 different integers in the Pop ID column? or just let all individuals to be mixed?

Besides that, what's your suggestion about run times and iterations. I was planning to follow Evanno's (2005) method to select the best K, using 20 iterations for each K, 10000 burnin, 10000 MCMC after burn in. Do I need to do something else after that? Or is it just about running the simulation again for the fixed K?

Thanks a lot!

--
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.

[zeamne T]

Dear Ana,

May i suggest you do few structure runs (e.g. 2 ), check for alpha convergence as Vikram suggested as each value of K (then you can fix the number of burn-ins). If you are using the GUI version of structure, they will plot it for you i think. Increase the burn-in if necessary. Then you can do a few iterations to check how long you need to collect data for (e.g. 5 runs), to see if the results are consistent across the 5 runs. If they are not consistent then you may have to increase the no. mcmc. Once you have decided on your run parameters, you can run the full analysis and estimate K as you mentioned.

Cheers,

YC

On 20 May 2015 at 10:56, Ana Mondon <anam...@gmail.com> wrote:

Thanks for your answer! As I can't do nested analysis what I did was assigning each of the 12 subsets a number of population (do you think that might hepl?)

For deltaK is selected K=1 to 15 (my number of subsets + 3, as suggested by Evanno et al), Burn-in= 10000 and 10000 MCMC, 20 iterations for each K. As I said, I chose those parameters based on Evanno et al and their method. Not sure if I should run more burn-in processes and MCMC, which I plan to do once I have selected K.

Is there anything I migth be missing? Steps to follow? 

Again, thanks a lot!

El martes, 19 de mayo de 2015, 16:41:11 (UTC-3), Vikram Chhatre escribió:

You can't do a nested analysis.  But if you find that structure is weak with 4 pops, you can always subset the data set and test further for that.

You may be able to get away with 10K burnin, but look at alpha progress to make sure it has converged prior to collecting samples.  I would collect data (mcmc) for at least 50K and preferably 100K. 

If you plan to use deltaK method, doing more Ks than you expect is useful, since the method will omit first and last K tested.

V

On Tue, May 19, 2015 at 3:02 PM, Ana Mondon <anam...@gmail.com> wrote:

Hello everyone. I'm starting to use STRUCTURE and have a few basic questions. My data set has 4 sampling locations, and individuals "split" into the same biotypes in each location. How do I specify this to the program? Should I use 12 different integers in the Pop ID column? or just let all individuals to be mixed?

Besides that, what's your suggestion about run times and iterations. I was planning to follow Evanno's (2005) method to select the best K, using 20 iterations for each K, 10000 burnin, 10000 MCMC after burn in. Do I need to do something else after that? Or is it just about running the simulation again for the fixed K?

Thanks a lot!

--
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.

--
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.

--

yujie

[Ana Mondon]

Dear YC, would you mind explaining to me a little bit more about this first step you are talking about? I mean the Alpha convergence. As you can imagine, I'm new with the program and just thought my first step was finding K, in the case of the DeltaK method, by running several similations. I'm using the GUI version of the program. Thanks a lot!

El miércoles, 20 de mayo de 2015, 1:12:50 (UTC-3), YC Tay escribió:

Dear Ana,

May i suggest you do few structure runs (e.g. 2 ), check for alpha convergence as Vikram suggested as each value of K (then you can fix the number of burn-ins). If you are using the GUI version of structure, they will plot it for you i think. Increase the burn-in if necessary. Then you can do a few iterations to check how long you need to collect data for (e.g. 5 runs), to see if the results are consistent across the 5 runs. If they are not consistent then you may have to increase the no. mcmc. Once you have decided on your run parameters, you can run the full analysis and estimate K as you mentioned.

Cheers,

YC

On 20 May 2015 at 10:56, Ana Mondon <anam...@gmail.com> wrote:

Thanks for your answer! As I can't do nested analysis what I did was assigning each of the 12 subsets a number of population (do you think that might hepl?)

For deltaK is selected K=1 to 15 (my number of subsets + 3, as suggested by Evanno et al), Burn-in= 10000 and 10000 MCMC, 20 iterations for each K. As I said, I chose those parameters based on Evanno et al and their method. Not sure if I should run more burn-in processes and MCMC, which I plan to do once I have selected K.

Is there anything I migth be missing? Steps to follow? 

Again, thanks a lot!

El martes, 19 de mayo de 2015, 16:41:11 (UTC-3), Vikram Chhatre escribió:

You can't do a nested analysis.  But if you find that structure is weak with 4 pops, you can always subset the data set and test further for that.

You may be able to get away with 10K burnin, but look at alpha progress to make sure it has converged prior to collecting samples.  I would collect data (mcmc) for at least 50K and preferably 100K. 

If you plan to use deltaK method, doing more Ks than you expect is useful, since the method will omit first and last K tested.

V

On Tue, May 19, 2015 at 3:02 PM, Ana Mondon <anam...@gmail.com> wrote:

Hello everyone. I'm starting to use STRUCTURE and have a few basic questions. My data set has 4 sampling locations, and individuals "split" into the same biotypes in each location. How do I specify this to the program? Should I use 12 different integers in the Pop ID column? or just let all individuals to be mixed?

Besides that, what's your suggestion about run times and iterations. I was planning to follow Evanno's (2005) method to select the best K, using 20 iterations for each K, 10000 burnin, 10000 MCMC after burn in. Do I need to do something else after that? Or is it just about running the simulation again for the fixed K?

Thanks a lot!

--
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.

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.

--

yujie

[Vikram Chhatre]

Ana,

MCMC is an algorithm to sample probabilities, in this case of inferring cluster memberships of individuals in your data set.  You are trying to run the mcmc chain long enough that the initial starting point has no bearing on the probability outcome.  

At some point during the run, the chain gets fully randomized (also called convergence).  This convergence can be assessed from parameters of the chain such as alpha.  In the front end, you will see a plot of alpha (yellow background, blue progress line IIRC).  Alpha should stabilize to a small distribution of values when it has converged.  If it's still jumping around a after 10000 sweeps, you need to do more burnin.

Burnin and MCMC are both names for the mcmc sweeps, the only difference being that you discard (burnin) samples prior to convergence and keep (mcmc) everything afterwards. 

As Yujie suggested, you could do some exploratory runs first and then set up your final run with as many sweeps as you can computationally afford.

Best

V

On Wed, May 20, 2015 at 9:22 AM, Ana Mondon <anam...@gmail.com> wrote:

Dear YC, would you mind explaining to me a little bit more about this first step you are talking about? I mean the Alpha convergence. As you can imagine, I'm new with the program and just thought my first step was finding K, in the case of the DeltaK method, by running several similations. I'm using the GUI version of the program. Thanks a lot!

El miércoles, 20 de mayo de 2015, 1:12:50 (UTC-3), YC Tay escribió:

Dear Ana,

May i suggest you do few structure runs (e.g. 2 ), check for alpha convergence as Vikram suggested as each value of K (then you can fix the number of burn-ins). If you are using the GUI version of structure, they will plot it for you i think. Increase the burn-in if necessary. Then you can do a few iterations to check how long you need to collect data for (e.g. 5 runs), to see if the results are consistent across the 5 runs. If they are not consistent then you may have to increase the no. mcmc. Once you have decided on your run parameters, you can run the full analysis and estimate K as you mentioned.

Cheers,

YC

On 20 May 2015 at 10:56, Ana Mondon <anam...@gmail.com> wrote:

Thanks for your answer! As I can't do nested analysis what I did was assigning each of the 12 subsets a number of population (do you think that might hepl?)

For deltaK is selected K=1 to 15 (my number of subsets + 3, as suggested by Evanno et al), Burn-in= 10000 and 10000 MCMC, 20 iterations for each K. As I said, I chose those parameters based on Evanno et al and their method. Not sure if I should run more burn-in processes and MCMC, which I plan to do once I have selected K.

Is there anything I migth be missing? Steps to follow? 

Again, thanks a lot!

El martes, 19 de mayo de 2015, 16:41:11 (UTC-3), Vikram Chhatre escribió:

You can't do a nested analysis.  But if you find that structure is weak with 4 pops, you can always subset the data set and test further for that.

You may be able to get away with 10K burnin, but look at alpha progress to make sure it has converged prior to collecting samples.  I would collect data (mcmc) for at least 50K and preferably 100K. 

If you plan to use deltaK method, doing more Ks than you expect is useful, since the method will omit first and last K tested.

V

On Tue, May 19, 2015 at 3:02 PM, Ana Mondon <anam...@gmail.com> wrote:

Hello everyone. I'm starting to use STRUCTURE and have a few basic questions. My data set has 4 sampling locations, and individuals "split" into the same biotypes in each location. How do I specify this to the program? Should I use 12 different integers in the Pop ID column? or just let all individuals to be mixed?

Besides that, what's your suggestion about run times and iterations. I was planning to follow Evanno's (2005) method to select the best K, using 20 iterations for each K, 10000 burnin, 10000 MCMC after burn in. Do I need to do something else after that? Or is it just about running the simulation again for the fixed K?

Thanks a lot!

--
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.

--
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.

--

yujie

--

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.