why no bootstrap support for internal node?

[Andrea Berardi]

Dear all,

I apologize if this a completely naive question, but I am wondering why there is no bootstrap support for all of the internal nodes in my small gene tree analyses. I have four taxa, species A, E, S, and I. Species I is my outgroup, and I typically run it as an outgroup in the RAxML command. I've been running both nucleotide sequences and amino acid sequences, always a single gene.

My particular question is why there is no bootstrap support for the internal nodes indicated by the blue circle in the attached PDF?

Is it because there are too few taxa? Or, am I misunderstanding something more basic about bootstrapping?

Tree 1 command:

raxmlHPC -p 12345 -m GTRGAMMA ­­HKY85 -s Genetree1.phy -o speciesI -n MYBFLcdsHKYboot -f a -x 12345 -#100

Tree 2 command:

raxmlHPC -p 12345 -m GTRCAT -s Genetree2.phy -o speciesI -n CNL_cds_HKY -f a -x 12345 -#100

Tree 3 command:

raxmlHPC-PTHREADS-AVX -T 16 -f a -m PROTGAMMAAUTO -p 12345 -x 12345 -#100 -s AAtree.fasta  -n aa_tree_for_astral

Any help would be greatly appreciated.

Thanks very much,

Andrea

[Lucas Czech]

Hi Andrea,

have you had a look at the output files of your RAxML analyses? It might be that the support values are in those files, but somehow not shown by the tree viewer that you are using for visualization. The missing values are the ones next to the root, which might be omitted by certain viewers.

So, please have a look at those files and/or post one of them here. Also, which tree viewer are you using?

Lastly, you might also be interested in the paper https://academic.oup.com/mbe/article/34/6/1535/3077051
where we critically review tree viewers with respect to issues arising in the context of support values in Newick files.

So long
Lucas

--
You received this message because you are subscribed to the Google Groups "raxml" group.
To unsubscribe from this group and stop receiving emails from it, send an email to raxml+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Grimm]

Hi Andrea,

"Or, am I misunderstanding something more basic about bootstrapping?"

Yes, you are (but not alone). Like nearly all tree inference methods, RAxML infers unrooted trees, so does all bootstrapping. The outgroup is only use to root the tree after inference and then the bootstrap support is mapped on the corresponding branches which define non-trivial taxon bipartitions (i.e. are no tips/terminal branches).

With four taxa, you have two possible unrooted trees with each one taxon bipartion

Topology 1 with A + E | I + S

Topology 2 with A + I | E + S

Bootstrapping now resamples your matrix, and generates a tree with either topology 1 or 2, i.e. showing either the taxon bipartition A + E vs I +S or A + I vs. E +S. Let's say topology 1 is the one found optimal by RAxML, then RAxML counts how often taxon bipartition 1 occurred in the bootstrap replicate trees.

In your case all bootstrap pseudoreplicates for gene one show topology/bipartition 1, and gene two for topology/bipartition 2. So actually your first tree shows something fundamentally different from the second and third. Gene tree 1 is fully incongruent to gene tree 2 and the AA tree.

You can't get support for the root in the post-analysis rooted tree, because the taxon bipartition defining it, I vs. A + E + S is a so-called trivial one, it will be found always in all trees (like the other terminals: A vs. E + I + S; E vs. A + I + S, and S vs. A + E + I)

Cheers, Guido

[Grimm]

Little graphical add-on: here's the potential unrooted trees (I forget the third alternative: A + S | E + I) for your four taxa that RAxML may find and counts to put a number on an according branch in the outgroup-rooted tree.

[Lucas Czech]

Oh right, I should have paid more attention... There is of course only one internal branch in the tree - which is the only one where a support value makes sense.

Thanks, Guido, for pointing this out!

Lucas

On 08/31/2018 02:44 PM, Grimm wrote:

Little graphical add-on: here's the potential unrooted trees (I forget the third alternative: A + S | E + I) for your four taxa that RAxML may find and counts to put a number on an according branch in the outgroup-rooted tree.

[Andrea Berardi]

Dear Guido and Lucas,

Thanks so much for your quick replies. Guido, the drawings of the unrooted trees really helped a lot, and brought back the memory of learning it a while ago. A bit embarrassing for me (oops) but thanks so much for explaining it kindly and clearly!

Lucas, your paper is very interesting - I had been using FigTree but it is good to know that there are limitations to the different tree viewers.

I will see if I can add another taxon in the future to at least increase the number of unrooted trees.

Thanks again!

Andrea

[Grimm]

Hi Andrea,

"I will see if I can add another taxon in the future to at least increase the number of unrooted trees."

If needed

This depends on your question and how sure you are that I is a good outgroup to root the AES tree.

If your question is just who of the three is sister to the other two, you would rather add more gene trees to see which of the two current alternatives is more favorable, both of which are unambiguous on their respective data sets. Given the relationship between the length of the tips and the internal branch, it may be at this point just a question of signal strength, but not gene tree congruence.

If you question is whether I is a valid outgroup for the AES tree, you need a fifth, second outgroup taxon X to test how well is the support for

X + I vs. A + E + S.

Cheers, Guido

[Andrea Berardi]

Dear Guido,

Thanks again. My goal is mostly to determine the relationship between A + E + S (and I now have about 1000 genes to test this with), but we have never really explored the validity of our I outgroup because it was the only availble option we had for some time. I may pursue an additional outgroup just to see, but I am glad to hear that my current approach seems appropriate for the relationship between A + E + S.

Thanks so much again for your help!

[Grimm]

Dear Andrea,

if you suspect some sort of reticulation in your three-plus-one problem, you may give PhyloNet (https://bioinfocs.rice.edu/phylonet/) a try with such kind of data.

E.g. if some of your 1000 genes stably (with consistent high BS support) prefer a different bipartition than the others, it could be tree-incompatible signal due to incomplete lineage sorting (no reticulation) or actual reticulation: introgression/(ancient) hybridisation. PhyloNet tests for this, will give you either a tree or an explicit phylogenetic network, and you can run it using your RAxML gene trees as input or directly from your data matrix.

Cheers, Guido