I want to construct a phylogenetic tree for a set of related proteins that are similar in structure but that lack significant sequence similarity. I ran a structure based alignment of the amino acid sequences through MrBayes and got an unrooted tree in which the clade credibility values for some branches are good (90% to 100%) and for others poor (between 50% and 90%).
My purpose was to justify dividing the proteins into separate families… so I intended to show only a clade credibility tree (with the probability for each clade on the diagram). My question is… do I show the entire tree with the bad values included… Or should I break it up and show only the good branches as separate trees?
As a separate question, would it be wrong for me to show a phylogram (with branch lengths) for this data set… because some of the sections are separated by more than 1 expected change per site?
Finally, MrBayes outputs an unrooted tree. Is it fair to display this as a midpoint rooted tree (by finding the longest tip to tip path length, setting it to be the root, and redrawing the tree by hand) or will this change the result in a way that I do not appreciate?
Thanks for help with any of these questions.
Chris
Washington University School of Medicine in Saint Louis.
Try BEAST2 and also *BEAST(starBEAST2) http://beast2.org/ BEAST runs Bayesian analyses and estimates rooted phylogenies without an outgroup, and can also help identify taxonomic groupings (eg. species trees) in *BEAST. You can then also use BP&P http://abacus.gene.ucl.ac.uk/software.html
Dear Christopher,
I have a few points
1) For phylogenetic analysis the sequences being compared should be homologous (common ancestral origin). The sequences you are comparing lacks significant similarity (this can happen in divergent sequences of same origin),.....Are the really homologous or.....are they analogous sequences evolved in a convergent manner for similar structures. In protein such examples do exists and these should be avoided for phylogenetic analysis
2) During statistical significance of clusters by methods such as 'Bootstrapping etc.' Even scores between 50 to 90 % are fine. Below 50% are generally considered low. If your credibility values are similar in nature...you may consider a cutoff of 50% or above as fine. many programs include option to select a cut off to display these scores...you may also do that...Even in very best trees..there are clusters/sub-clusters with bad statistical scores and it is normal....as we know this is just one gene/protein tree and not each polymorphic site evolves in a similar manner...there can be some unexpected problems also...
3) If you just want to separate the major protein families...an uprooted tree may also suffice the requirement...However, if you plan to discuss their evolution or its direction then it may not be a good option...
4) In the sample set if you know the most divergent sequence (out group) you may choose it as a root of the tree and this will resolve many problems in explaining the dataset
5) Branch lengths if really required may be displayed...or else may be substituted by a common distance scale
bye
Hello Christopher,
If all you are doing is defining gene families and you are not trying to trace the common ancestry of the gene families to one another then a unrooted tree is fine. If you do want a rooted tree and do not have an outgroup then as Jim pointed out BEAST will be your best option since it uses a coalescent approach to determine the root. *BEAST is not appropriate for your data since it is a species tree method that calculates a species tree from multiple gene trees, thus you need several independent loci to run this analysis and is not what you are doing here.
If you are building an amino acid tree and homology can not be assessed with the DNA sequence you will have to give a justification for how you know the proteins are related outside of just having similar function or else reviewers will rip you apart just like Ajay pointed out.
As far as support values go you can collapse the phylogeny into polytomies if the bpp are below a certain threshold (many use 0.50 or 0.70), or you can just show the whole phylogeny with the all support values or symbols for high verse low support. This is more of an aesthetics issue then a statistic issues, just make sure to note the poor supports and not make any conclusions off of the groupings with these low support values.
Hope this helps.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology