Dear Ioannis

In principle, you are not allowed to change results - e.g. to drop out what you can't explain or what does not look how you expected... You are only allowed to edit your results for better visualization. Results always means something and you have to explain them biologically!

Let me go step by step:

- You have multiple sequences per species and thus multiple terminal branches (what you call tips) per species. First question is: Are all individuals of the morphological determined species in the same monophyletic clade?

For monophyletic clade go to. https://en.wikipedia.org/wiki/Clade

Are all individuals of the species descendants of the same ancestor? Like the red and blue clades in wikipedia?

If not, then you have paraphyly and things became complicated...

If yes, next question: Have terminal branches a high support (>70% bootstrap values)? Then, different individuals of the same species contain genetic variability on the locus you used. Don't drop out this information but comment it. It could be that you used individuals from a wide geographic range. Then genetic variability is not surprising.

- Let say all sequences of a species form a monophyletic clade and subclades are not well supported. Then, I would not consider the subclade branches because they are statistically not relevant.

Then, you can collapse subclade branches but do it systematically (and automatically) and not randomly and subjective tip by tip! I would collapse, for example, by a node support

I am sorry for the upload error ;))

Dear Gabrielle and Carolina,

Thank you both very much for you answers.

Gabrielle, I am able to extract a picture of the collapsed tree, but I need it as a tree file, for the phylogenetic comparative analysis I want to do later.

Carolina, the clades that I was able to collapse were indeed monophyletic. I would prefer to use collapsed clades instead of randomly removing terminal branches, but with the R package I am using (ggtree) I was not able to save the tree with the collapsed clades as a new tree object.

The reason I want to have a single terminal clade per species is that I am concerned that if I have for example 50 terminal clades for one species and only a couple for another, this over-representation of the first one could affect the results of my future analysis, e.g. ancestral state reconstruction. Do you believe this is valid concern?

If I am to go back to the collapsed clades approach, is the software you suggested able to save the new tree with the collapsed clades as a new tree object?

Again, thank you both very much for your answers. I will be looking forward to any further suggestions.

I seems I had the same problem with multiple answers. :)

Hi Ioannis

"Do you believe this is valid concern?"

Yes!

But then we are speaking from different things. Collapsing branches is a more "aesthetic" option to visualize your tree topology better AFTER you performed the analysis. But what you need is a more balanced data set from the beginning. That means not to overload a single species with a lot of records, while others are underrepresented in comparison to the first one. This can produce bias in tree reconstruction. I would reduce the amount of specimens per species, for example, to 1-2 representatives of different geographic provenances or morphological features that may be in question. If the 50 have exactly the same sequence it does not make sense to repeat them because then they are identical independently of the geographic origin or morphological property... Phylogenetic speaking: they don't contain parsimony information.

How to get a data set that includes only those sequences that contain SNPs?

1) Select them in your alignment by similarity. Most alignment programs have this option. Then delete what ever is identical.

2) You can load your dataset in RAxML. These software will write in the log file "warnings" for all identical sequences (because they burden the computing power) and it produces a new input file containing only sequences with SNPs. You can run RAxML using R, or you use the CIPRES online suite. Use RAxML-Blackbox that is pretty fast. Download the new input file and use it for additional analyses.

3) Use SplitsTree4. It has the option File > Tools > Group identical haplotypes. Then export the file as nexus or fasta.

Attention: If you have gaps in your alignment, all programs will keep those sequences containing gaps because they can be consider a fifth state or as missing information... Even if all other sites of the sequence with a gap are identical with other sequences.

Finally, you are using R that is a brilliant solution for all kind of statistics. But I ask you again which kind of algorithm are you applying for the construction of the tree? Probabilistic such as maximum likelihood or Bayesian analysis? Or genetic distances like neighbor joining or UPMG? All these approaches may have a big impact on the resulting tree topology. Then how are you assessing the accuracy of the tree topology? Bootstrapping?

Did you just load your sequences in R and then you run an algorithm that was recommended to you? But you don't know which substitution model it uses...

I don't know how important is the tree for your work but I want to recommend you to apply the state of the art for an ancestral character state reconstruction: 1) perform an alignment in an alignment program (Clustal, MEGA...), check the alignment by eye, are there regions containing gaps, ambiguous sites or a lot of identical haplotypes? 2) Once the alignment is "clean" (this data set is the basis for all additional analyses! It must be clean), export it as fasta, nexus or phylip. 3) Use the nexus or phylip for the assessment of the substitution model (software jModeltest) and/or apply a maximum likelihood analysis (RAxML, PAUP, PhyML...) and/or a Baysian analysis (MrBayes or BEAST) directly, applying the GTR substitution model. All these programs export the tree in newick format that can be used for an ancestral state reconstruction. Be aware a newick file also contains the bootstrapping values or posterior probabilities (Bayesian). Check the tree topology in programs that visualize trees: FigTree or TreeGraph...

More about newick format: http://evolution.genetics.washington.edu/phylip/newicktree.html

I hope I could help you (it is not easy to produce a phylogeny by the state of the art and I cannot explain you every thing that is necessary. Sorry).

Best regards

Carolina

Dear Carolina,

Thank you very much for your input. You provided me with a lot of useful bits of information which made me think a lot about the next steps of my analysis.

I ended up with multiple, different sequences per species because, in order to get my sequences, I use ReproPhylo, a phylogenomics pipeline, which allows me to download every available sequence for a specific gene for all my species of interest, in an automated way. On one hand it saves a lot of time, on the other hand you can end up with some species being over-represented in the final tree.

As for the sequence alignment, I use MAFFT.

I am in an initial stage, so I do maximum likelihood analysis with RAxML, which is faster, but once I figure out how to get a tree with a single terminal clade per species, I will also do Baysian analysis with MrBayes.

Again, thank you very much for your reply.

Kind regards,

Ioannis

Carolina's advice is technically sound, but I am not sure that there is any philosophical difference between pruning terminals off the tree after the analysis, and deciding to exclude taxa you already have gathered data for prior to running the analysis. That is not to say that either is necessarily bad, and people routinely do both in many studies published in MP&E, etc.

I would say that as long as you explicitly describe your methodology and rationale for including or excluding taxa, it does not really make a lot of difference whether you do it before or after. If you describe your methods in a way that somebody could repeat the analysis, your approach will be much more transparent than what most people do, which is to sweep problems like this under the rug. Besides, you can always put a tree with all the data in the supplementary material.

Hi Andrew. Thank you for your answer. Would the time of exclusion of taxa affect any future analysis? I plan to do ancestral state reconstruction.

The more data include in the analysis, the more robust the inference of ancestral states will be. If it were my project, I would start with as many terminals as possible and only reduce if absolutely necessary.

You may want to try species tree reconstruction method like *BEAST and run the subsequent analyses using the species tree. You can include as many individuals as possible in the species tree analysis; the resulted tree will always be one tip per species.

For ancestral state reconstruction, if you are interested in "reconstructing the ancestral state," (estimating the transition matrix Q) then it does not matter whether you have multiple or single representatives per species. That is, only anagenetic evolution was accounted for.

However, if you are taking cladogenetic events into account (such as biogeographic state reconstruction using DEC or all the State-dependent speciation extinction [SSE] models), then your result will be significantly affected. By including different number of individuals, the frequency of lineage split (cladogenesis) will be affected, and so the estimated rates of speciation & extinction.

I agree with Jen-Pan, BEAST 2 (http://www.beast2.org/) allows you to include as many sequences as you want and always have an output grouping together the sequences belonging to the same group (e.g. species).

Moreover, BEAST allows you to include multiple alignments (different genes) without having to concatenate them.

Kyra suggest you contact Christine Lambkin at the Queensland museum for advice. Tell her you are working on fireflies. [email protected]

Thank you Lesley A. Ballantyne . Will update you as soon as I can coz it's been a while.

Hope this is useful. Most of the Pyrophanes are in the Philippines. Look forward to hearing how you are going.

How Max Score, Total score, Query coverage, E value and a Max ident for sequences are used to decide which of those sequences to add to my alignment and thus to include on tree?