Partition Finder and Bayes

[Gina Filloramo]

Hi Rob, 

I've been experimenting using Partition Finder with my multi-gene phylogenetic dataset. The manual is very user friendly and I've been able to get some analyses up and running. I do have a question about implementing the best partitioning scheme for  Bayesian analysis. I currently am using the MrBayes plugin for Geneious 6.1.6. For one particular analysis, the best model for all of the subsets except for one was GTR+I+G. The one subset with a the different model called for HKY+I+G. I'm curious as to how I would specify in my Bayes Block that particular partitions are under one model and other partitions are under another. I've copied my best scheme here: 

Best partitioning scheme

Scheme Name       : step_3

Scheme lnL        : -113371.16985

Scheme BIC        : 229297.141859

Number of params  : 289

Number of sites   : 6906

Number of subsets : 9

Subset | Best Model | Subset Partitions              | Subset Sites                   | Alignment                               

1      | GTR+I+G    | Gene1_pos1, Gene1_pos2, Gene1_pos3 | 1-2627\3, 2-2627\3, 3-2627\3   | 

2      | GTR+I+G    | Gene2_pos1                     | 2628-3859\3                    | 

3      | GTR+I+G    | Gene2_pos2, Gene4_pos2         | 2629-3859\3, 5545-6906\3       | 

4      | HKY+I+G    | Gene2_pos3                     | 2630-3859\3                    | 

5      | GTR+I+G    | Gene3_pos1                     | 3860-5543\3                    | 

6      | GTR+I+G    | Gene3_pos2                     | 3861-5543\3                    | 

7      | GTR+I+G    | Gene3_pos3                     | 3862-5543\3                    | 

8      | GTR+I+G    | Gene4_pos1                     | 5544-6906\3                    | 

9      | GTR+I+G    | Gene4_pos3                     | 5546-6906\3                    | 

Any insight into appropriately incorporating different models based on partition would be appreciated. Thanks again for designing this program! It is very helpful!

Best, 

Gina

[Rob Lanfear]

Hi Gina,

Thanks for getting in touch. 

Before answering your question, I want to clarify a very common misunderstanding that you might have made (it's not totally clear from what you wrote, but just in case). PartitionFinder has identified 9 different models for your data, not two. What's true is that 8 of these models have the same name (GTR+I+G), however, they do not have the same parameters, otherwise PartitionFinder would have grouped them together. It's important to have this distinction clear before you start trying to set up your Bayesian analysis.

To figure out how to set up a partitioned analysis in MrBayes take a look at teh wiki:

http://mrbayes.sourceforge.net/wiki/index.php/Analyzing_a_Partitioned_Data_Set_3.2

and also at Brant Faircloth's gist, which tells you how to specify different models (e.g. HKY, GTR) in MrBayes:

https://gist.github.com/brantfaircloth/895282

However, note that the MrBayes wiki is not very complete on the issue of partitioned models. You'll need to unlink parameters across your partitions, and allow the overall rate of evolution to vary among partitions. To do that, learn about the following commands:

prset ratepr=variable;

unlink shape=(all) pinvar=(all) statefreq=(all) revmat=(all);

In case it helps, I've pasted below a mrbayes block I've used for a partitioned analysis in the past (in this case, all models were GTR+I+G, but hopefully it helps).

Cheers,

Rob

begin mrbayes;

outgroup ZEA_MAYS;

[Codon Partitions for Coding Sequences]

charset matK1st=598-2137\3;

charset matK2nd=597-2737\3;

charset matK3rd=596-2737\3;

charset phyB1st=2138-3477\3;

charset phyB2nd=2139-3477\3;

charset phyB3rd=2140-3477\3;

charset ndhF1st=4636-6767\3;

charset ndhF2nd=4635-6767\3;

charset ndhF3rd=4634-6767\3;

charset rbcL1st=6768-8194\3;

charset rbcL2nd=6769-8194\3;

charset rbcL3rd=6770-8194\3;

[Non-Coding Partitions]

charset ITS = 1-595;

charset rpl16 = 3478-4633;

charset trnLF = 8195-8988;

[Partitions]

partition combined = 15: matK1st, matK2nd, matK3rd, phyB1st, phyB2nd, phyB3rd, ndhF1st, ndhF2nd, ndhF3rd, rbcL1st, rbcL2nd, rbcL3rd, ITS, rpl16, trnLF;

set partition= combined;

[set model to GTR+I+G for every partition]

Prset applyto = (all) statefreqpr=dirichlet(1,1,1,1);

lset applyto=(all) nst=6 rates=invgamma;

[unlink all parameters for all partitions]

prset ratepr=variable;

unlink shape=(all) pinvar=(all) statefreq=(all) revmat=(all);

[set up mcmc]

mcmc ngen =  100000000 printfreq = 500  samplefreq = 10000 nchains = 4 savebrlens = yes nruns=2;

end;

[Gina Filloramo]

Hi Rob, 

The links were very helpful! I figured out how to set up my partitioned analysis in Geneious using a Bayes block with different models for my partitions :)

Thanks so much for your help!
Gina