High-throughput data dimension reduction via seeded canonical correlation analysis

Abstract

Canonical correlation analysis (CCA) is one of popular statistical methodologies in multivariate analysis, especially, instudying relation of two sets of variables. However, if sample sizes are smaller than the maximum of the dimensionsof two sets of variables, it is not plausible to construct canonical coefficient matrices due to failure of inverting samplecovariance matrices. In this article, we develop a two step procedure of CCA implemented in such situation. For this,seeded dimension reduction is adapted into CCA. Numerical studies confirm the approach, and two real data analysesare presented.