Implement support of phylogenetic trees to calculate genetic distances between direct sister species

[mvences]

In order to improve the informativeness of the TaxI3 output, one important additional features would be the support for phylogenetic trees in order to identify among the different species the direct sister species. This will be an extension of the "all against all" comparison

It will include three levels of complexity.

The first level will be support for phylogenetic trees provided as additional input by the user. In these, we need to make sure that the taxon names in the tree can be directly linked to the sequence files. We should first implement the feature with tab-delimited and Excel input - here the requirement will be that the taxon name in the tree must be identical with the seqid field. I will below provide some suggestions for implementing this first level. Once it is done, we can think about the second and third level.

The second will be to make TaxI3 calculate trees on its own, directly from the sequences. There are many programs for this, we can start implementing very simple algorithms but eventually want to have some more complex ones. Some preliminary suggestions on this below.

The third level will be to also consider in the calculation some kind of support value. Whether two taxa are sister to each other in a phylogenetic tree can be weakly or strongly supported by the data, and there are means to assess this, for instance through pseudoreplication (bootstrap) which is implemented in many tree-searching programs. This we will leave until the very end.

For being able to work with phylogenetic trees, we need a dedicated library for trees. There are several of them:

• The Phylo module of Biopython also some basic manipulation of trees
• ETE3 is quite advanced and allows for a large number of options: http://etetoolkit.org/
• Dendropy is an alternative to ETE3. Personally I would prefer Dendropy for one specific reason: it is quite well documented, and it was developed by Jeet Sukumaran who will become a collaborator in the project, so it will be easy to adjust the library if we think it is necessary. So, unless ETE3 has some really important complex features missing in Dendropy, I would prefer Dendropy.

What is a sister species? See the graph below for a very simple explanation.

However, in a data set for TaxI3, we will typically have trees with more than one sequence per species. So, in the following example, A and B are sister species. An important concept for this is that all sequences assigned in the data set to one species are grouped together (they are monophyletic):

While in the the following example, sequences A vs. B are not reciprocally monophyletic, and so cannot be considered as true sister species:

So here is what the program should do:

• Read a tree inputted by the user (in the so-called Newick format)
• Identify from the tree which species are true sister species (category 1), and which ones have all sequences in one monophyletic group not including sequences of any other species, but without being reciprocally monophyletic (category 2).
• For all distance calculations, tables and graphs, provide information for these sister species separately from those of other intra-species comparisons. So (if requested by the user) there would be comparisons of (1) all intra-species comparisons, (2) sister species (depending on user input, also including category-2 species pairs), (3) all other intra-species comparisons within the same genus and (4) all inter-species comparisons of the same genus.
• Identify all non-monophyletic species and output a list of them
• Make these calculations optional, that is, provide the user with a checkbox to select this additional analysis or not [remember that at some point later, we will completely overhaul the GUI, so for now, the GUI can just be extended and the additional options added somewhere, no need to spend much time on designing this for now]

Lastly, here already some preliminary thoughts about how to calculate trees within TaxI (second level of complexity, see above):

• Biopython has a very simple Neighbor-joining algorithm in the Bio.Phylo.TreeConstruction module
• Eventually we would like some maximum likelihood tree algorithm. The good and fast and scalable ones of course are not written in Python. There are some in Perl (http://fastml.tau.ac.il/source.php) or in C (https://github.com/stamatak/ExaML). I would prefer using ExaML (or RAxML) which are under continuous development and have become much of a standard. But this would mean to wrap a complex C code into TaxI3 which I am not sure will be easy!

[necrosovereign]

@mvences Could you provide sample input for this functionality?

[necrosovereign]

I also need a table of sequences that would be inputted with these trees.

[mvences]

OK, the following ZIP file includes:

• an example of a tab file with sequences (not aligned)
• a tree derived from these sequences, with seqid values labelling the branches (the tree is provided in two variants, with or without branch length information; both should be "understood" and correctly interpreted by the program. TaxI3_example_with_tree.zip

The interpretation of the tree and sequences should be as in the following graphs. There are two pairs of sister species as recognizable from the trees. The genetic distance values from these two comparisons should be tabulated separately from the other inter-species values.