Bayesian Clustering Approaches for Discrete Data

Unsupervised classification or clustering uses no a priori knowledge of the labels of the observations in the process of categorizing data. The research contained in this thesis focuses on the machine learning of discrete-valued gene expression datasets using clustering, with the aim of identifying gene co-expression networks. Specifically, a number of topics surrounding the use of mixture models and Markov chain Monte Carlo (MCMC) methods in clustering of discrete data from high-throughput transcriptome sequencing technologies is presented. After outlining current challenges and gaps in research with respect to clustering approaches, three mixture model-based clustering methods are presented: mixtures of multivariate Poisson-log normal distributions, mixtures of multivariate Poisson-log normal factor analyzers and mixtures of matrix-variate Poisson-log normal distributions. Significance, innovation, limitations and a number of future directions stemming from this research are discussed.

Grantor dc:publisher

University of Guelph

Year dc:date.issued

2017

Author dc:creator

• Silva, H. Anjali

Advisor dc:contributor.advisor

• Rothstein, Steven

• Clustering
• RNA sequencing
• Discrete data
• Multivariate Poisson-Log Normal distribution
• Markov chain Monte Carlo
• Factor analyzers
• Matrix variate distribution
• Co-expression network

Language dc:language.iso

en

Handle dc:identifier.uri

http://hdl.handle.net/10214/13025

OAI record oai:identifier

https://atrium.lib.uoguelph.ca/server/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Aatrium.lib.uoguelph.ca%3A10214%2F13025