Re: Shaking up the bird family tree

[David Marjanovic <david.marjanovic@gmx.at>]

> Another way to put the original rationale for the Metaves would be 
> "presently supported by at most one unambiguous apomorphy in a character 
> type where extreme convergent evolution is known". And it is probably not 
> a good idea to make much of such "clades" either (Would anyone here 
> confidently write a 16-pages paper proposing clades based on a *morph* 
> analysis yielding such a situation?).

Yes. Happens all the time.

Especially because you have misunderstood what "unambiguous synapomorphy" 
means. Look at this tree:

--+--A 0
 `--+--B 0
    `--+--C ?
       `--+--D 1
          `--E 1

Here, state 1 is an ambiguous synapomorphy of D and E because we don't know 
if it's actually an autapomorphy of a larger clade.

--+--A 1
 `--+--B 0
    `--+--C 0
       `--+--D 0
          `--+--E 1
             `--F 1

Here, state 1 is an unambiguous synapomorphy of E and F. That it also 
appears in A does not matter. "Unambiguous" does not mean "unique".

> It is strong simply because it's whole blocks of identical bases missing 
> or inserted in a single instant, overruling single-base changes across the 
> entire sequence.

No, that's not how maximum-likelihood analysis is done. Instead, there's a 
"gap-opening cost" and a "gap extension cost" that you have to choose; an 
indel of length 2 is given higher weight than an indel of length 1, but much 
less than twice that weight.

> OTOH, there is sufficient evidence that certain loci "attract" 
> transposons: the "21bp deletion" may just as well be a widespread 21bp 
> insertion (you cannot tell in the scope of a cladistic analysis).

Isn't 21 bp way too short for a transposon? Surely transposase has more 
than, like, five amino acids?!? And why should a transposon insert into the 
exact same place again and again and again?

> Changing the alignment algorithm (in BLAST terms: using PAM and not 
> BLOSUM) would probably prevent you from ever finding this supposed 
> dichotomy. Alternatively, one might treat indels as a single character 
> each - which they technically *are*. In that case too, the signal would be 
> easily overwhelmed.

Someone should test that (by setting the gap extension cost to 0).

> In conclusion: we need bones (or analyses thereof), because the bases 
> simply won't cut.

Bones will certainly help, and they are the only thing that can tell us 
where to put the fossils. But analyzing them is so much work... The paper 
we're talking about had pure calculation times of several months, but the 
whole thing took maybe a year or two. The work by Livezey & Zusi took ten 
years, didn't it, and it still contains way too few fossils...