several partitions

24 views
Skip to first unread message

andrea nino

unread,
Apr 27, 2021, 7:03:17 PMApr 27
to PartitionFinder
Dear Rob, 
I used a strategy by partitioned my dataset (13500 pb * 150 species) by gen and by codon to Protein coding Genes, and one by each rRNA with model_selection : bic and search: greedy. All was ok with the partitionFinder run, and Iqtree, Mrbayes, RaxML analysis. However, to my Beast analysis (using the same partition scheme and 7 calibration points) I cannot arrive to optimize my ESS values after several N° of generations. I supposed it could  arrive one day running several analysis and many hr in a cluster. Do you believe that optimize the number of generation could help to optimize the Beast analysis? 
What could be the best strategy? Grouping the partition by gen, or by position? or by position and same model ?  Or this does not have any sense.
Or relented a new scheme partition in PartitionFinder?

Thank for the advises 

Best regards, 
Andrea

Subset | Best Model | # sites    | subset id                        | Partition names                                                                                     
1      | GTR+I+G    | 777        | aa6d08870474a261954d433ad287322a | 12S                                                                                                 
2      | GTR+I+G    | 1137       | c770cc808857d12eb2d67da7e0758b9b | 16S                                                                                                 
3      | GTR+I+G    | 1112       | 2ef2eb84dc48341275890c9886e931cd | COII_pos1, ATP6_pos1, COIII_pos1, CYTB_pos1                                                        
4      | TVM+I+G    | 1014       | 3bd306a9701fb1001b9b06f8625942f2 | COI_pos2, COIII_pos2, ATP6_pos2                                                                    
5      | TVM+G      | 423        | 14af0a2a5e4771d8fcec37f54c19c4fa | ND3_pos3, ATP6_pos3, ATP8_pos3                                                                     
6      | HKY+I+G    | 70         | 09c06b283a6ba155b862fbbb38eeaae7 | ATP8_pos1                                                                                          
7      | GTR+I+G    | 283        | 604c00155ccf4a0be8e7dbd1ec888904 | ND6_pos2, ATP8_pos2                                                                                 
8      | GTR+I+G    | 517        | 63e10bda2d363c9dee5e81713c85b4ae | COI_pos1                                                                                           
9      | GTR+I+G    | 752        | 2b5edf5b945982971c25c94bf9a8fc9f | COI_pos3, COII_pos3                                                                                 
10     | GTR+I+G    | 615        | 97a01e0ce67a8045a0482aad57a39f0e | COII_pos2, CYTB_pos2                                                                               
11     | TIM+G      | 265        | e98e1a37d5da6b8c54cd5707f0eafde4 | COIII_pos3                                                                                         
12     | TIM+I+G    | 380        | ed77f53d9829f38e52853f1e541a6764 | CYTB_pos3                                                                                          
13     | GTR+I+G    | 1455       | cc75992d6172e44b09acff9311e955e3 | ND5_pos1, ND1_pos1, ND4L_pos1, ND4_pos1                                                            
14     | GTR+I+G    | 314        | 837d34d1fd73ea9fed0c069125ab885a | ND1_pos2                                                                                            
15     | TVM+I+G    | 314        | 719cf8dbcbc87def839e14b6f74e2304 | ND1_pos3                                                                                            
16     | GTR+I+G    | 374        | 1f9895b348e888b7f719089ffb41e425 | ND2_pos1                                                                                           
17     | TVM+I+G    | 495        | 7b4514749caa8db90ae3ae7e6e46972b | ND2_pos2, ND3_pos2                                                                                 
18     | GTR+G      | 374        | c1c9c0bacf26d0bd81b6f430703637e1 | ND2_pos3                                                                                           
19     | GTR+I+G    | 334        | 643fb50e0a0911a120c27a203a85bdef | ND6_pos1, ND3_pos1                                                                                 
20     | GTR+I+G    | 1141       | 4e0dde062f976253d07b25928fb79cea | ND4L_pos2, ND4_pos2, ND5_pos2                                                                      
21     | TIM+I+G    | 101        | d59867060c15e1b8554ccb066da1647e | ND4L_pos3                                                                                           
22     | TRN+I+G    | 1040       | 276ad8fb996a1e06214eedd85c87439a | ND5_pos3, ND4_pos3                                                                                 
23     | HKY+G      | 213        | 15994501a9ddd135de599d0ba39dd87c | ND6_pos3                                                                                            


Rob Lanfear

unread,
Apr 27, 2021, 7:08:02 PMApr 27
to PartitionFinder
Hi Andrea,

There are almost limitless reasons why your BEAST analyses may not be converging. I'd be somewhat surprised if the partitioning scheme was the culprit, though it's possible. 

One issue is that PartitionFinder is not built for a Bayesian framework (it's very explicitly built for a frequentist framework). So there's no guarantee that the partitioning scheme you get are appropriate for a Bayesian analysis. 

The usual solution to low ESS values is very simple - run the analysis for longer, and/or run a series of independent chains and combine the output. Given long enough runs you will usually get good sampling. More generally, there are lots of ways to tune an MCMC to help sample properly from the posterior distribution, and lots of options to do this in BEAST. That's a topic for the BEAST discussion group though. 

Rob

Yu Sun

unread,
May 3, 2021, 10:37:26 PMMay 3
to partiti...@googlegroups.com
Hi Nino,
    I also used the same strategy to get the results and to run the Bayesian, it took me a long time to run, how long have you used it?  Thank you.

Best,
Yu

andrea nino <nerak...@gmail.com> 于2021年4月27日周二 下午6:03写道:
--
You received this message because you are subscribed to the Google Groups "PartitionFinder" group.
To unsubscribe from this group and stop receiving emails from it, send an email to partitionfind...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/partitionfinder/7b1ecc6e-e2cc-4d63-8ace-a6d9bb429a50n%40googlegroups.com.

andrea nino

unread,
May 6, 2021, 6:09:11 AMMay 6
to partiti...@googlegroups.com
Dear Rob, 
thank you so much for your answer. Yes, I agree that this partition scheme is correct and totally possible to do a Beast analysis. I will try to run a larger N° generation and long runs and run a series of independent chains.

Best,

Andrea

--
You received this message because you are subscribed to a topic in the Google Groups "PartitionFinder" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/partitionfinder/yGpXWIw-2wQ/unsubscribe.
To unsubscribe from this group and all its topics, send an email to partitionfind...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/partitionfinder/7ce67119-ef84-434b-85d7-17d5e42d2f37n%40googlegroups.com.

andrea nino

unread,
May 6, 2021, 6:16:54 AMMay 6
to partiti...@googlegroups.com
Hi Yu,
I did 6 independent runs with 50 million of generation, sampling log each 1000 gn. Each one takes around 90hr on the cipres server. 
Could you indicate to me how long time runs your analisis? and how many generations? 
Thank you for the information.

Best 
Andrea 

Yu Sun

unread,
May 16, 2021, 12:11:28 AMMay 16
to partiti...@googlegroups.com
I ran it for a long time until my computer crashed... btw, the generations I used 10 million. So, I decided to run it on PhyloBayes.

Yu

andrea nino <nerak...@gmail.com> 于2021年5月6日周四 上午5:16写道:
Reply all
Reply to author
Forward
0 new messages