Concerned that a larger Cau et al. 2015 theropod tree did not match the theropod subset of the large reptile tree, yesterday we looked at a reduced Cau et al. 2015 theropod taxon list that matched taxa to the large reptile tree. This was done to delete potentially incomplete taxa that might reduce resolution. Yesterday only 518 of their 1549 characters were employed, but these covered all parts from head to toe. By comparison, the large reptile tree employes only 228 characters, of which only 151 are taxonomically informative to the theropod subset.
Earlier phylogenetic studies
(Wiens 2003) demonstrated only marginal advances in accuracy with any character or taxon list over 150 as the logarithm curves toward and plateaus at a line parallel to the X-axis.
Today we’ll add back
those missing 1000+ characters to the Cau et all (2015) study to match their 1549 character total. Against the expectations and conclusions of Wiens 2003, I’m hoping for more resolution in the Cau et al. tree with the addition of 1000+ theropod specific characters.
Still of great concern,
the Cau et al trees still do not have greater resolution and higher Bootstrap scores with 3x as many characters vs. those 518 characters tested yesterday. By comparison the Cau et al tree employs 10x more parsimony informative characters employed by the large reptile tree (1549 vs 151). And yet the large reptile tree found complete resolution.
the Cau et al and large reptile tree topologies still do not match each other in every detail when pruned to a similar taxon list (Figs. 1,2). I still wonder, why do these differences exist? The usual answer: “taxon inclusion/exclusion,” cannot apply in this case. Inaccurate scoring on one or both matrices is now a stronger possibility that will have to be explored.
The theropod subset of the large reptile tree,
employing only 228 traits, remains fully resolved with a different topology (Fig. 2). The high bootstrap figures tell you that the tree topology remains largely fully resolved even when higher tree lengths are permitted. Smaller numbers appear at weaker nodes. None of the Bootstrap scores are in the 50s or 60s and only a few are in the 70s.
In future blog posts
I will attempt to dig deeper into this quandary, trying to figure out why the tree topologies still differ and why there is less resolution in the Cau, Brougham and Naish tree despite the higher and more specific character count. The devil is obviously in the details. I have no idea what I will eventually find here. I have already triple-checked the large reptile tree scores and reconstructions. Cau et al. did not provide evidence that they created reconstructions or examined every bone for every score they used. Rather, if they employed a prior dataset, they may have TRUSTED the data therein. I don’t look forward to checking every score in the Cau et al. tree. That will take a rainy weekend, at least based on its great size. But I think it has to be done.
Thanks to Andrea Cau
for responding to my requests for data and characters. I will try to treat all aspects of this problem fairly and seriously without animosity.
Cau A, Brougham T and Naish D. 2015. The Phylogenetic Affinities of the Bizarre Late Cretaceous Romanian Theropod Balaur bondoc (Dinosauria, Maniraptora): Dromaeosaurid or Flightless Bird? PeerJ. 3: E1032. DOI: dx.doi.org/10.7717/peerj.1032
Wiens JJ 2003. Missing data, incomplete taxa, and phylogenetic accuracy. Systematic Biology 52: 528–538.