Signature discovery for personalized medicine

Various types of genome-wide data, such as sequence and gene expression data, have been generated and are available from public databases. These genome-wide data present major computational challenges as the number of variables far exceeds the number of observations. Many computational tools have been developed for the analyses of these high dimensional data, and these methods have led to improved understanding of molecular biology. In particular, signature discovery (also known as variable selection or feature selection), a machine learning technique in which subsets of variables are selected to build robust models, are useful in mining these high-dimensional functional genomic data. In this paper, we will review the applications of signature discovery methods in mining these high dimensional data. Specifically, we will focus on two applications, namely, the identification of signature genes predictive of disease phenotypes and the inference of regulatory networks. Signature genes predictive of disease phenotypes can be potentially used in the diagnosis and prognosis of diseases. Regulatory networks that capture the gene-to-gene influences can be used to provide the context of therapeutic intervention.