Model-Based Clustering of DNA Methylation Array Data

Abstract

Clustering refers to the "grouping" of observations into a discrete set of classes, such that observations in the same class are more similar compared to objects between classes. In the context of DNA methylation data, clustering can be used to discover novel molecular subtypes or to identify biological pathways comprised of co-methylated CpG dinucleotides, depending on whether the samples or the CpGs themselves are being clustered. In this chapter, we focus on the problem of clustering samples/subjects on the basis of their methylation profile. We begin by discussing the motivation behind clustering DNA methylation data, the nature of DNA methylation data generated from the Illumina BeadArrays, and three promising model-based clustering methods. In addition to providing a methodological overview of each of the three methods, we also demonstrate their application using a publicly available data set deposited in the Gene Expression Omnibus (GEO) database. Issues such as feature selection and comparison of clustering partitions will also be discussed.