MichelleKendall
commented
6 years ago Hi Thomas,
Thanks for getting in touch. It's great to hear that the treeConcordance idea could be useful for you and I'm sorry that's it's unsuitable for your analysis as it stands. We are in the process of redrafting that paper / code and all suggestions are very welcome.
Thanks for your detailed example and yes you're absolutely right about what's going on. I can see that the function is working as intended but I agree that the emphasis on root position is giving a low score for Ind Tree 2 and obscuring the fact that a major clade is in full agreement with the category tree. It is difficult within a single score measure to capture all the possible tree differences which people may or may not care about but I will have a think about how we could handle this better. In the meantime, some ideas:
- How important / precious is your rooting method? Do you care at all that Ind Tree 2 is rooted differently or could you, as part of your workflow, root all trees so that the red category is basal? This would be appropriate in some situations and not others - it really depends on your tree inference method and whether you have strong beliefs about the red category being an outgroup.
- Could you include a clear outgroup in your workflow, so that all trees effectively get rooted the same way? (Again, not always appropriate.)
- If you were to examine random subtrees of your tree pairs, recalculate the concordance, and then find an average per individual tree, I believe you would find that the score for Ind Tree 1 would remain at 1 and the score for Ind Tree 2 would be raised, reflecting the full agreement on a major clade. Might that be a useful measure, or too hard to interpret?
- I'm assuming you have many Ind Trees to examine which might have other differences you want to capture. You may find that although the relative difference between 1 and 0.2 is very large, the ranking of the individual trees from "most similar" to "least similar" to the category tree agrees with your intuition, even if the magnitude of the scores does not. I'd be interested to hear if that's the case.
I hope that helps. Do please let me know if you have any further queries or ideas because I am keen to make this measure as useful as possible.
Cheers,
Michelle
thackl
.
Hi,
I'm trying to use treespace for the comparison of individual gene phylogenies to overall "species" phylogenies. Most of my data is from single-cell sequencing, which means individual genomes and gene sets are incomplete, and therefore tree leaf sets between trees are not identical. I was pretty excited when I read your tutorial on "Comparing trees by tip label categories" because it looks like an approach that could help me a lot with my analyses.
However, I came across an issue when trying to use treeConcordance to compare gene trees to the overall phylogeny (see figure below). The Category tree represents the expected topology for certain subclades, the two individual trees represent trees obtained from single gene families. Tree 1 is in absolute agreement with the reference tree, and also gets a treeConcordance of 1. Tree 2 is not identical, though at least naively I would say quite similar to the reference tree. However, treeConcordance is only 0.2.
After reading the tutorial more carefully and looking at the source code, I think I understand why: the "same placement" of a clade in the categorial tree and the individual tree is determined by the depth of the clade-MRCA in the tree (correct me if I am wrong). So because in Tree 2, the most basal clade is split into two clades, an additional depth level is introduced, MRCA-depths of all other clades are shifted, and they don't contribute to concordance anymore.
I guess this is the expected behaviour of the treeConcordance, although it seems counterintuitive. Unfortunately, it also makes treeConcordance unsuitable for the type of comparison I would like to do. Do you have any idea for alternative measurements, workaround or other methods that I could use in my situation?
Cheers Thomas