Issue 748 in beast-mcmc: TreeAnnotator: Sum of clade credibility score for tree reported wrongly when burnin is == 0

[beast...@googlecode.com]

Status: New
Owner: ----
Labels: Type-Defect Priority-Medium

New issue 748 by jeetsukumaran: TreeAnnotator: Sum of clade credibility
score for tree reported wrongly when burnin is == 0
https://code.google.com/p/beast-mcmc/issues/detail?id=748

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).





--
You received this message because this project is configured to send all
issue notifications to this address.
You may adjust your notification preferences at:
https://code.google.com/hosting/settings

[beast...@googlecode.com]

Comment #1 on issue 748 by ramb...@gmail.com: TreeAnnotator: Sum of clade

credibility score for tree reported wrongly when burnin is == 0
https://code.google.com/p/beast-mcmc/issues/detail?id=748

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).

[beast...@googlecode.com]

Updates:
Status: WontFix

Comment #2 on issue 748 by ramb...@gmail.com: TreeAnnotator: Sum of clade

credibility score for tree reported wrongly when burnin is == 0
https://code.google.com/p/beast-mcmc/issues/detail?id=748

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