TreeAnnotator: Sum of clade credibility score for tree reported wrongly when burnin is == 0

[GoogleCodeExporter]

This is regarding v1.8.1

What steps will reproduce the problem?

1. Run TreeAnnotator, selecting "Maximum Sum of Clade Credibilities" as the 
target tree 
2. With burnin > 0 (in number of trees), score is correct
3. With burnin == 0, score is incorrect

In (2), above, the score is calculated by summing posteriors over all internal 
nodes.

In (3), above, the score is calculated by summing posteriors over *all* nodes, 
including terminals.

The (albeit, dated) documentation implies that (2) is correct. 

Incidentally, there is a lot of confusing and contradicting information 
regarding what the following two trees should be called, both within the BEAST 
suite as well as in supporting documentation.

I.   the tree with the maximum product of clade posteriors
II.  the tree with the maximum sum of clade posteriors

The programs of the current (v1.8.1) BEAST suite, as Heled & Bouckart (2013) 
[doi:10.1186/1471-2148-13-221] use the term "maximum clade credibility tree" 
for Tree Type (I).

However, the BEAST wiki, Wikipedia, as well as the BEAST documentation bundled 
with BEAST v1.8.1 define the "maximum clade credibility tree" as Tree Type 
(II), i.e., the tree with the maximum sum of clade posteriors rather tan the 
maximum product of clade posteriors. The first two sources use the term 
"maximum credibility tree" to refer to Tree Type (I).

I am implementing support for calculating these scores and selecting maximum 
credibility trees of both types (I) and (II) in DendroPy. I would like to go 
with established terminology, but it seems difficult to pin down. Currently I 
am leaning toward using the cumbersome yet explicit (and more general) 
"product_of_split_support_score" and "sum_of_split_support_score" for 
internals. Given that there is a peer-reviewed publication backing the term 
"maximum clade credibility tree" for Tree Type (I), I might stick with that, 
and use "maximum sum of clade credibilities tree" for Tree Type (II). 

Original issue reported on code.google.com by jeetsukumaran on 30 Sep 2014 at 8:42

[GoogleCodeExporter]

Maximum Clade Credibility (MCC) tree is _de facto_ used in the literature to 
mean the tree with the maximum product of the clade credibilities. I don't see 
any justification for using the maximum sum of the clade credibilities and it 
is present only for historical reasons. I propose removing this option in 
future versions. 'Maximum Credibility' tree would seem symantically the same as 
the 'Maximum A Posteriori' tree (i.e., the modal topology). 

Original comment by ramb...@gmail.com on 8 Dec 2014 at 2:16

[GoogleCodeExporter]

Marking this as WontFix as I have removed sum of clade credibility from 
TreeAnnotator.

Original comment by ramb...@gmail.com on 5 Mar 2015 at 9:27

• Changed state: WontFix