Optimize best tree from Phyml by Garli

[Yuan Jin]

Hi all,

I am trying to optimize tree topology and branch length generated from Phyml.

Should I just provide the tree topology for start or a tree with branch length?

I am also wondering whether I should use the model parameters estimated from Phyml or

let Garli to re-estimate them? Thanks in advance.

Best regards

Yuan Jin

[Derrick Zwickl]

Hi Yuan Jin,

Assuming that you are using version 2.0, you should use the "optimizeinputonly" setting, described here:
https://www.nescent.org/wg_garli/GARLI_Configuration_Settings#optimizeinputonly_.28do_not_search.2C_only_optimize_model_and_branch_lengths_on_user_trees.29
Note that that option requires a NEXUS format input tree though.

It shouldn't matter whether or not it has branch lengths, and the optimized branch lengths returned by GARLI should be the same regardless.  You should probably let GARLI reoptimize model parameters, as some of the models are defined slightly differently in the two programs.

If your goal in doing this reoptimization is to get likelihood scores, keep in mind that scores are not necessarily comparable between programs, so only compare likelihood values for different tree obtained by optimizing branchlengths and model parameters in a single program.

Let me know if you have other questions.
Best,
Derrick

[Yuan Jin]

Hi,Derek

Thanks for your reply.I am using version 2.0, yet I didn't notice the newly added option "optimizeinputonly" . I checked the related link and found it seems the option won't optimized the tree topology. I've tried run Garli with the model parameters unset, and treated the best tree searched by PhyML as a user tree. Should I better use the "optimizeinputonly"  option to optimize the banchlength only?

All the best

Yuan Jin

在 2013年9月12日星期四UTC+8上午5时34分29秒，Derrick Zwickl写道：

[Derrick Zwickl]

Hi Juan Jin,

I'm sorry, I guess that I misunderstood your initial question.  My initial answer assumed that you were just trying to optimize branch lengths and model parameters on the tree that Phyml inferred for you.  If you'd like to use the Phyml tree as a starting point and try to find better trees, just run the program normally with the file containing that tree specified as the streefname setting.  Note that if GARLI returns a different tree (with a better likelihood score), it doesn't necessarily mean that this tree was missed by PHYML, because under PHYML's slightly different model the likelihood might not be better.

Let me know if you have other questions.

Derrick

--
You received this message because you are subscribed to the Google Groups "garli_users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to garli_users...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.