A Likelihood-Based Method for Clustering Pathogen Variants with Multiple-Trait Data

• Activity Number: 33
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Ziqian Zhou*+ and Michael Kosoy and Kung-Sik Chan
• Companies: University of Iowa and CDC and University of Iowa
• Keywords: Clustering ; Classification ; Mixture model ; Bioinformatics ; Phylogeny ; EM algorithm

Abstract

There are no universally acceptable ways to classify the variants of a pathogen. Traditionally, classification can be done by clustering pathogen variants using data from a single trait, e.g. DNA sequence, biochemical traits or host species information. Combining multiple traits for clustering the variants of a pathogen has not been done due to lack of appropriate methodologies. Here we propose a novel approach for clustering the variants of a pathogen based on multiple traits. The new framework assumes a phylogenic tree sub-model for each trait with the tree realizations constrained to be of a particular form suitable for clustering and that the traits are conditionally independent given the true cluster. We developed an efficient algorithm to maximize the likelihood (posterior) via the EM algorithm. The new approach is illustrated with simulated data and is also applied to cluster bacteria Bartonella specimens based on DNA sequences, geographical locations and host species.