Assisted clustering of gene expression data using ANCut

Sebastian J. Teran Hidalgo, Mengyun Wu, Shuangge Ma

In biomedical research, gene expression profiling studies have been extensively conducted. The analysis of gene expression data has led to a deeper understanding of human genetics as well as practically useful models. Clustering analysis has been a critical component of gene expression data analysis and can reveal the (previously unknown) interconnections among genes. With the high dimensionality of gene expression data, many of the existing clustering methods and results are not as satisfactory. Intuitively, this is caused by "a lack of information". In recent profiling studies, a prominent trend is to collect data on gene expressions as well as their regulators (copy number alteration, microRNA, methylation, etc.) on the same subjects, making it possible to borrow information from other types of omics measurements in gene expression analysis. In this study, an ANCut approach is developed, which is built on the regularized estimation and NCut techniques. An effective R code that implements this approach is developed. Simulation shows that the proposed approach outperforms direct competitors. The analysis of TCGA (The Cancer Genome Atlas) data further demonstrates its satisfactory performance. We propose a more effective clustering analysis of gene expression data, with the assistance of information from regulators. It provides a new venue for analyzing gene expression data based on the assisted analysis strategy.