Concatenation methods have been criticized for creating inflated support values, and ignoring potentially multiple independent gene histories in your data... Thus it is possible to create apparently strong support for an "incorrect" topology, or misrepresent the support for a "correct" topology (insofar as there exists an ascertainable "correct" topology).

In addition to checking out SNAPP as already suggested, I would look into SVDQuartets, which runs MUCH faster than SNAPP. SNAPP (in my experience) can take months even on HPC for a large dataset.

Hi Ian, 

Both PartitionFinder and jModelTest have no problems analyzing entire non-partitioned data sets. These programs only perform best-fit model selection for the partitions you specify beforehand. For PartitionFinder, you can specify a single partition in the Data Blocks that will include all sites in your sequence, e.g.:

Charset = 1 - 650000;  // no need to divide it by 1

For jModelTest, you can just load your entire sequence file making sure to convert it into phylip format; then you can do model selection for the entire data set.

I would also suggest that you consider using SNAPP add-on for BEAST 2 software that's designed for that specific type of data.

Cheers,

Oksana

Hi Oksana,

Fantastic, thank you very much for explaining it all so clearly! I'll have a look at BEAST 2 and the SNAPP add-on now.

Cheers,

Ian

Concatenation methods have been criticized for creating inflated support values, and ignoring potentially multiple independent gene histories in your data... Thus it is possible to create apparently strong support for an "incorrect" topology, or misrepresent the support for a "correct" topology (insofar as there exists an ascertainable "correct" topology).

In addition to checking out SNAPP as already suggested, I would look into SVDQuartets, which runs MUCH faster than SNAPP. SNAPP (in my experience) can take months even on HPC for a large dataset.

Hello Ian,

I would also suggest you read the following paper about acquisition bias correction in inferring SNP phylogenies.

http://sysbio.oxfordjournals.org/content/early/2015/09/15/sysbio.syv053.long

Also I will point out that, while this is still a point of contention in the literature, using linked SNPs can lead to bias in the resulting phylogeny if there is gene discordance. Most species tree methods are made using unlinked SNPs and thus using highly linked SNPs can be problematic. Below is a good article that sums up some of these issues.

http://www.sciencedirect.com/science/article/pii/S1055790315003309

Best Regards

Thanks for the advice and input Tyler and Max! Apologies for the slow response, I thought I'd replied. Max, thank you for including those paper links too, I appreciate it.

Best wishes,

Ian