Unsupervised Machine Learning to Support Functional Characterization of Genes: Emphasis on Cluster Description and Class Discovery

In recent years, multiple types of high-throughput functional genomic data have become available to facilitate rapid functional annotation of sequenced genomes. However, such data often sacrifice specificity for scale, and thus sophisticated analysis methods are necessary to make accurate predictions of gene function based on large-scale datasets. This chapter presents an overview of unsupervised analysis of microarray data followed by an in-depth discussion of integrated analysis of heterogeneous biological data for accurate gene function prediction. This discussion focuses on a general probabilistic method for such integration, called MAGIC, and provides an overview of the methodology, application and evaluation of this technology.