Re: Juravenator; or How not to perform a phylogenetic analysis

[Michael Mortimer <mickey_mortimer111@msn.com>]

Today's description of Juravenator included a phylogenetic analysis, which 
moved me to write this post.  I noted a couple problems which plague a large 
portion of published phylogenetic analyses I see.

Gohlich and Chiappe's analysis is the Theropod Working Group (TWG) analysis 
with 27 taxa excluded, 8 taxa added, 43 characters deleted and 10 characters 
added.  WHY did they delete taxa and characters!?  Studies show that 
including taxa and characters is VERY important to resolving "correct" 
trees.  I mean, I know it takes time to code taxa, but they were already 
coded!  It took more time to bother deleting them.  Ughh.  Maybe they 
weren't relevent taxa, you say.  Gohlich and Chiappe were trying to 
determine basal coelurosaur phylogeny, not phylogeny of all non-pygostylian 
coelurosaurs after all.  Alas, the deleted taxa include Alvarezsaurus and 
Pelecanimimus, which as basal representatives of their clades, are very 
important when placing taxa falling near Ornithomimosauria and 
Alvarezsauridae.  And why bother deleting any at all?  They're not going to 
harm the analysis if included.

They recovered a tree where the compsognathid/coelurid taxa formed a 
polytomy at the base of Maniraptora.

What to do when you have a large polytomy?  Well, here's what not to do.  
Gohlich and Chiappe ran a second analysis "omitting seven taxa, either less 
relevant (i.e., Alxasaurus elesitaiensis, Erlikosaurus andrewsi) or largely 
incomplete (i.e., Mirischia asymmetrica, Scipionyx samniticus, Nqwebasaurus 
thwazi, Coelurus fragilis, Nedcolbertia justinhoffmanni)."  I already 
explained that taxa should not be deleted, especially those close to your 
group of interest.  How are therizinosaurs less relevant than, say... 
oviraptorosaurs?  Or avialans?  Also, something that almost every author 
needs to learn... *ahem* incomplete taxa don't necessarily lead to decreased 
resolution.  In fact, they can actually help.  It all depends on what 
codings are unknown.  So don't delete incomplete taxa just because they're 
incomplete.  They may influence topology regardless.

After deleting the seven taxa, Gohlich and Chiappe recovered new trees.  
Ornitholestes was still a basal maniraptoran, but Compsognathus, 
Sinosauropteryx, Huaxiagnathus and Juravenator ended up in a monophyletic 
clade sister to Maniraptoriformes.  The fact this topology is incongruent 
with the original topology should have raised alarm bells.

So here's what you should do if you have a large polytomy you want resolved. 
 Display a majority rule consensus in PAUP (contree "" majrule = yes) to 
find the poorly resolved taxa (they'll have the most seldomly occuring nodes 
around them).  Then delete one taxon at a time BUT don't delete them from 
the analysis.  Instead, just delete them from the saved trees.  When you 
type "delete Juravenator" for instance, PAUP will ask whether you want to 
delete it from the saved trees or just prune it.  Select prune.  This will 
leave the trees the same, but without your deleted taxon and any polytomies 
caused by it falling out in different positions.  Keep deleting taxa until 
your tree is as resolved as you want.  If this is done to Gohlich and 
Chiappe's tree, the following tree results-

|--non-maniraptoriformes
`--+--Ornithomimosauria
  `--+--Juravenator
     |--Scipionyx
     `--+--Compsognathus
        `--+--Huaxiagnathus
           |--Sinosauropteryx
           `--+--Ornitholestes
              `--"oviraptoriformes"+paravians

The excluded taxa help move 'compsognathids' into Maniraptora and make them 
paraphyletic to derived maniraptorans.  Juravenator is no longer a 
compsognathid, but is instead more basal.  And it's SO easy to do.  So 
please, authors of future papers...
1. Don't exclude taxa from analyses because they are incomplete.
2. If you want to resolve a polytomy, just prune the poorly resolved taxa 
from the tree.  Don't delete them and rerun the analysis.

Mickey Mortimer