Do you have a strong feeling about supertree methods versus concatenation? We have some in prep work showing ASTRAL outperformed some of the others you listed (MP-EST, BUCKy, and *BEAST). How divergent are the species you are interested in comparing? I usually identify homologs with OrthoMCL, refine orthology with single-gene trees and PhyloTreePruner, and then concatenate with FASconCAT and analyze a partitioned supermatrix.

Dear Max,

Get the following paper: http://www.biomedcentral.com/1741-7007/10/13

All of these thechnics are founded on the calculation and the comparison between statistical "distances" that exist between different samples. Then, the two major questions are: (i) what is the definition of these "distances"? and (ii) what are the algorithms that have been used to obtain the supertrees in these softwares? The subsidiary question is: what is the result that you (reasonnably) expect and the subsidiary +++ question is: are your samples adapted to one or the other of theses approaches?…

Regards

Christian

I have tried several approaches and programs with ddRAD. My conclusion, if you don't have anything that can confound simple rule "older divergence - bigger genetic distance" all methods will agree about topology (but may disagree in estimation of time, effective population size etc. if the specific program can do this). In my case, I failed with BUCKy since individual trees from orthologs were badly resolved due to few information in short reads. Concatenated sequences with ambiguities produced box structures in networks (analyses with Phangorn and then SplitsTree if I remember correctly) but this is indirect indication of hybridization and in general nobody knows how properly deal with heterozygous positions. SNAPP did a good job with SNP and I consider coalescent approach in building multilocus trees as a correct one, but again, it had not notice when hybrids were added to dataset and place them next to one of parental species. Moreover, it is not possible to calculate more than 12-15 samples in reasonable time. SVD-quartet by Kubatko can do it much faster, but does not estimate time and population size. And I am still looking for some software which account for introgression and incomplete lineage sorting in the same time. Have heard something about TreeMix, FastSimCoal and other modelling aproaches etc. but never tried. Mankind will be grateful to the person who will try them all and report what is the best.

I have developed SISRS to estimate phylogenies from NGS data.

https://github.com/rachelss/SISRS/

 http://www.biomedcentral.com/1471-2105/16/193

SISRS outputs an alignment of orthologous variable sites without using a reference genome. Given the amount of data produced I recommend SVD-quartets or a RAxML analysis of concatenated data with a model that compensates for exclusion on invariable sites. The loci incorporating these sites produce near-perfect estimates of divergence dates.

It's actually a bit strange to me that you ask about NGS and species trees. Do you mean many aligned loci? phylogenomics? Why the NGS part of this question?

Hello All,

I am not specifically asking this question for my research exactly. Although, I am using ddRAD and making phlyogenies, so this discussion is of interest. I just feel that there is not enough literature out there on the pros and cons of different methods, although a lot more has been coming out recently. So, I though a discussion could be fun.

Kevin -

From experience I know that ASTRAL runs much faster than the other methods you mentioned, especially if you are using ASTRAL-II. It also seems to give some of the best trees in my very small experience and is my preferred method. I am interested to see your in prep work when it comes out. Also, I am interested in how you are quantifying performances (nodal support, branch lengths, run time).  

To answer your question on divergence times, my tree currently contains 16 species (although I am adding a few more species). Based on fossil calibration of mtDNA the genus is ~30-40mya old and the splits within my ingroup range from ~27mya and ~1mya. There are also several introgression, which makes the phylogeny interesting.

Oleksandr -

I too tried BUCKy, but since I have ~20,000 loci for 200 samples each loci 87bps, I got poor results and took months to run on just a subset of samples. Attached is a paper on fire finches and parasitic finches that looked at several methods including BUCKy, ASTRAL, MP-est, NJst, STAR, and STELLS, that found similar results with BUCKy. Have you ever tried STELLS, it looks interesting and I have not tried it yet.

Rachel -

I have not tried SISRS before, but it looks interesting. I used pyRAD for filtering, clustering, and aligning. I have also used concentrated data for making phylogenies with RAxML and Mr. Bayes and while this did work great for the most point, it could over inflate support values for anomaly zones or introgression, which is where trees using individual loci can be very useful.

Best Regards    

http://www.ncbi.nlm.nih.gov/pubmed/21873298 is a good paper on not inflating support values. I recommend support as % of genes supporting a node. Also use the SH test to test various hypotheses.

New paper that might be of interest:

Coalescent methods for estimating species trees from phylogenomic data

http://onlinelibrary.wiley.com.ezproxy.library.uq.edu.au/doi/10.1111/jse.12160/pdf

Some links are available here:

http://bioinformaticssoftwareandtools.co.in/