Bootstrap Consensus Trees - (Sep/05/2009 )

Hi all,

Does anyone know if it is better to use to the "original" tree or the "bootstrap consensus" tree of a NJ analyses in MEGA? They give slightly different topologies.

If I have understood my literature correctly, I should be using the consensus tree in the Maximum Parsimony analyses, but I wonder if I should be doing the same in NJ?

Thanks!

PS - the sequences I am looking at are all from one species, and are fairly well conserved.

You are talking about different statistical concepts. So when you are doing a Maximum Parsimony analyeses it starts with a different assumption than a neighbour joining. Which method to use depends on your data matrix. NJ is usually the faster method but also more prune to "mistakes" (i.e. wrong topology). MP gets slower the more data you want to look at. Bootstraping tells you how credible your nodes are - so how often they are found when re-sampling your tree. So you are not talking about independent analyese here. If your bootstrapvalues are low you should think of running another analyes.

Probably you should use Modeltest or a similar programm to find the best settings for your analyeses.

Hi gebirgsziege,

I purposefully chose to use two separate statistical analyses, so that I could confirm the overall results of the NJ tree with a more thorough MP tree. I intend on using NJ for day-to day typing, but I wanted to scrutinise my phylogeny protocol first before setting off on using the protocol for further research (ergo the MP tree to double check my initial results).

Where my confusion lies is that in the MEGA software, after the NJ analyses (w/1000 bootstraps) I am presented with the option to view the "original" tree, or a "bootstrap consensus" tree. So I am unsure whether or not I should be using the bootstrap consensus tree. The main clades have good bootstrap values in both versions of the tree, but some of the minor topology is different. (I just need to settle on one particular method to use in my subsaquent studies.)

PS - I tried the new version of ModelTest, but I could not get any sequence format to load onto the program...

gebirgsziege on Sep 7 2009, 04:25 PM said:

modeltest accepts the ".nex" (nexus) format. For converting data into nexus I use Mesquite (freeware) which I find a quite useful program for working with trees as well.

Bootstraping resamples your tree (if I got you right 1000x) and creates a new tree every time. Therefore the bootstrap consensus tree is the consensus tree of your re-samplings of your initial tree. It should be similar, but can be different from your initial tree. The more similar the trees of different analyses are (ML, MP, NJ, baysian) the more likely your tree is true.

You have to decide which method produces the "best" result for your assumtion or for the "true" phylogenetic position of your organism/gene (like viewed at in literature at the moment). And depending on the data volume you will have to process you should decide for speed of the algorithm (+settings) you choose.....

Ahh. I did not try a .nex file. Thanks , I;ll give it a go.

When I am don, I hope that the NJ trees will give me similar enough topographies and bootstrap values to the MP trees that I can just use the NJ trees for my regular typing. Fingers crossed!

gebirgsziege on Sep 8 2009, 04:50 PM said:

If I get you right you just want to confirm that seq A = seq B? Then NJ should be enough......although I think if this is what you attemt for, you could do with a similarity matrix....

Sort of.

We're developing a typing system for a new virus, so some evolutionary relationship information is needed. The NJ method I think should give us the right amount of information presented in a graphical way (as long as the bootstrap values are good anyways!)

gebirgsziege on Sep 8 2009, 06:38 PM said:

Unfortunately, Mesquite is a Mac program (using PC). I found an alternate site which converts files into interleaved phylip and nexus formats, but even those files will not work for me.
Any other programs that you know of that I can test out my sequences for models of substitution?

mesquite can be used on PC as well....you just have to check out the Win-version. I am using it on PC as well......

But I think Modeltest needs paup to run.....so maybe you can try one of the following sites for modeltesting:

findmodel

will post the second later...it is on my computer at home.

Mesquite website, also with download links for the windows version.


Here is a collection of model-selection/testing software.