Sparse Canonical Correlation Analysis for Biomarker Discovery: A Case Study in Tuberculosis

Biomarker discovery from’omics data is a challenging task due to the high dimensionality of data and the relative scarcity of samples. Here we explore the potential of canonical correlation analysis, a family of methods that finds correlated components in two views. In particular we use the recently introduced technique of sparse canonical correlation analysis that finds a projection directions that are primally sparse in one of the views and dually sparse in the other view. Our experiments show that the method is able to discover meaningful feature combinations that may have use as biomarkers for tuberculosis.