Hi! I am trying to build a phylogeny for 100+ Hemipteran taxa. I was able to make a super-matrix tree using 13 genes, but there was a problem with it. One of the subgroups should be monophyletic but it was coming out as paraphyletic. Eventually I was able to solve that using two different approaches.
The first was that checked the trees produced using each gene on its own and I examined which genes were giving the group as paraphyletic. After excluding these genes, I got the expected topology, but I lost 20+ taxa. Among the genes I had to remove, was CO1 and I am not sure if that's ok or not.
The second approach was that I created a constrain tree by hand which was separating my taxa in 3 groups; the outgroup, and the two main groups which were making up my sample of Hemipteran taxa. I left every other relationship on this tree as a polytomy, to be resolved by RAxML. These worked and gave a tree similar to the one I got using the previous approach, allowing me to keep all my taxa and the CO1 sequences.
Are these two approaches acceptable? Is one of them better than the other?
Thank you all for taking the time to read my question. Any feedback will be greatly appreciated!
First, I fully support constrained tree searches, those are great. Especially if some relationships such as well-established clades deeper in the tree are not the point of interest.
People have often identified outlier genes in this way, so that is not necessarily an issue. Also COI is pretty fast so maybe it is not surprising that information about deeper relationships is lost. But - since you lose 20 taxa that apparently only have COI in the supermatrix- the phylogenetic conflict could be a matter of taxon sampling too. Maybe take a quick look at these taxa and check that they are not early diverging lineages in the two ingroup hemipteran clades that might have some influence over this particular split.
I would suggest that you get the site likelihoods for both your constrained search and the original unconstrained search and check if the original unconstrained tree actually improves the likelihood significantly using a KH test. You would have to get them with -f G and then use CONSEL. RAxML does implement the SH test though with -h H, but the SH test is really meant for when you are comparing the ML topology with more than one other candidate topologies. My expectation is that you will fail to reject the hypothesis that the paraphyletic ML tree has a significantly greater likelihood score than the constrained tree, providing additional support for why the constrained search is appropriate in this case.
George Tiley thank you very much for the detailed answer and the suggestions!
Are there any publications regarding the use of constraint trees?
Sure - these are some empirical examples where constraint trees were used to test taxonomic hypotheses. And the Hemiptera one gives a bit of discussion on the use of constraint trees when doing these supermatrix searches that often incorporate a lot of taxa that might only have short barcoding sequences available. Note these are just a few recent examples that use constraints combined with topology testing, but there are a lot more out there.
Fisher AF, et al. 2016. Evolutionary history of chloridoid grasses estimated from 122 nuclear loci. Mol Phylogenet Evol. 105:1-14.
Percy DM, et al. 2018. Resolving the psyllid tree of life: phylogenomic analyses of the superfamily Psylloidea (Hemiptera). Syst Entomol. 43:762-776.
Liu SH, et al. 2018. Genome skimming provides new insight into the relationships in Ludwigia section Macrocarpon, a polyploid complex. Am J Bot. 105:875-877
In your particular case, I would include both approaches in the paper (and the unconstrained analysis too). The reasons are basically as George Tiley said – looking at the individual genes loses information by losing taxa, using a constraint tree loses information by not telling you why some genes fit a different tree better, so it's best to have both.
I routinely use constraint trees to test how much less certain hypotheses are supported than my unconstrained results.
- Badges
- Science topic
- Similar topics
- Condensed Matter Physics
- Zero