Biomarker discovery and visualization in gene expression data with efficient generalized matrix approximations

Visualizing microarray data for biomarker discovery by matrix reordering and replicator dynamics

In most microarray data sets, there are often multiple sample classes, which are categorized into the normal or diseased type. Traditional feature selection methods consider multiple classes equally without paying attention to the upregulation/downregulation across the normal and diseased classes; while the specific gene selection methods for biomarker discovery particularly consider differential gene expressions across the normal and diseased classes, but ignore the existence of multiple classes. More importantly, there are few visualization algorithms to assist biomarker discovery from microarray data. In this paper, to help users visually analyze microarray data and improve biomarker discovery, we propose to employ matrix reordering techniques that have been developed and used in matrix computation. In particular, we generalized a well-known population genetic algorithm, namely, replicator dynamics, to reorder a microarray data matrix with multiple classes. The new algorithm simultaneously takes into account the global between-class data pattern and local within-class data pattern. Our results showed that our matrix reordering algorithm not only provides a visualization method to effectively analyze microarray data on both genes and samples, but also improves the accuracy of classifying the samples.