[Evolution_RU] Journal Club in Evolution, Jan 26: Long Branch Attraction in Bayesian Analysis

Lena Struwe struwe at AESOP.Rutgers.edu
Wed Jan 20 14:54:04 EST 2010


Journal Club in Evolution for Spring 2010 starts on Tuesday Jan 26, 2 PM 
in 262 Foran Hall on Cook Campus, and the first subject will be a recent 
paper highlighted by Dana Price (thanks, Dana!).  We will discuss the 
pros and cons of different phylogenetic methods as it relates to the 
long branch attraction problem.

The paper is:
Long-Branch Attraction Bias and Inconsistency in Bayesian Phylogenetics, 
by Bryan Kolaczkowski & Joseph W. Thornton, 2009, PLoS ONE 4(12): e7891. 
doi:10.1371/journal.pone.0007891:
available here, 
http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0007891

Short background:  Long branch attraction (LBA) is a known problem in 
maximum parsimony analysis when you use highly divergent taxa, which can 
result in monophyletic clades in groups that are really paraphyletic. 
Maximum likelihood (ML) is less susceptible to this problem. This paper 
shows the LBA effect in Bayesian Analysis.

A good introduction to the LBA and the "Felsenstein zone" is in the 
lecture notes by Delwiche, available here:
http://www.life.umd.edu/labs/delwiche/MSyst/lec/Properties.html

Links: Wikipedia, LBA: http://en.wikipedia.org/wiki/Long_branch_attraction

Journal Club in Evolution is open to all students and faculty interested 
in organismal evolution at any level (individual to phyla, genes to 
morphology) and is arranged by Lena Struwe and Karl Kjer. To get all 
journal club announcements and other evolution-related information 
campus-wide, sign up for the Mailing list in Evolution at Rutgers: 
https://email.rutgers.edu/mailman/listinfo/evolution_ru

See you!

Lena



-------- Original Message --------
Subject: 	[Evolution_RU] LBA and bayes
Date: 	Thu, 07 Jan 2010 11:02:14 -0500
From: 	Dana Price <d.price at rutgers.edu>
To: 	evolution_ru at email.rutgers.edu



In a nutshell, bayes' ability to integrate over parameter uncertainty 
(i.e branch lengths) does counterintuitive things when your datasets get 
bigger and bigger.  The attached shows it to be hugely susceptible to 
long branch attraction with posteriors converging to 1.0 on incorrect 
trees, and a weakness of likelihood that may now be a strength.

Dana



-- 
*************************************
Dr. Lena Struwe
Associate Professor
Dept. of Ecology, Evolution, & Natural Resources/ Dept. of Plant Biology & Pathology
Rutgers University 
237 Foran Hall, 59 Dudley Road
New Brunswick, NJ 08901-8551, USA
phone: (732) 932-9711 ext. 235, fax: (732) 932-9441
e-mail: struwe at aesop.rutgers.edu  
home page: www.rci.rutgers.edu/~struwe    
GENTIAN RESEARCH NETWORK: http://gentian.rutgers.edu 
Director, CHRYSLER HERBARIUM: http://herbarium.rutgers.edu
 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://email.rutgers.edu/pipermail/evolution_ru/attachments/20100120/0ff98e30/attachment.html>


More information about the Evolution_RU mailing list