Campus Event Calendar
Event Entry
New for: D1, D2, D3, D4, D5
What and Who
Model Selection for Mixtures of Mutagenetic Trees
Junming Yin
IMPRS
Masters' Lunch
AG 1, AG 2, AG 3, AG 4, AG 5
MPI Audience
Note: We use this to send email in the morning.
Date, Time and Location
Sunday, 1 February 2004
12:59
60 Minutes
Warning: Date has changed!
46.1 - MPII
024
Saarbrücken
Abstract
Mutagenetic trees constitute a particular class of
restricted Bayesian tree models that have been designed to model
cumulative evolutionary processes. Mixture models of mutagenetic
trees have been used to model the evolution of drug resistance in
the HIV genome. There exists an EM-like algorithm for estimating a
K-component mutagenetic trees mixture model from data. Here we
address the model selection problem of identifying the correct
number of trees K. We present a modified Bayesian Information
Criterion (BIC) that includes an estimate of the structural
redundancy between the tree components. We compare our method to
standard BIC and to cross-validation on simulated and real HIV
data. We find that standard BIC tends to select evolutionary
models that contain redundancy. Both cross-validation and the
modified BIC perform significantly better, but the latter is more
practical for real world data sets.
Contact
Kerstin Kathy Meyer-Ross
93 25 226
--email hidden
passcode not visible
logged in users only
Kerstin Meyer-Ross, 11/27/2004 10:47
Kerstin Meyer-Ross, 11/25/2004 17:44 -- Created document.
Kerstin Meyer-Ross, 11/25/2004 17:44 -- Created document.