Identifying common and distinctive processes underlying multiset data

Abstract

In many research domains it has become a common practice to rely on multiple sources of data to study the same object of interest. Examples include a systems biology approach to immunology with collection of both gene expression data and immunological readouts for the same set of subjects, and the use of several high-throughput techniques for the same set of fermentation batches. A major challenge is to find the processes underlying such multiset data and to disentangle therein the common processes from those that are distinctive for a specific source. Several integrative methods have been proposed to address this challenge including canonical correlation analysis, simultaneous component analysis, OnPLS, generalized singular value decomposition, DISCO-SCA, and ECO-POWER. To get a better understanding 1) of the methods with respect to finding common and distinctive components and 2) of the relations between these methods, this paper brings the methods together and compares them both on a theoretical level and in terms of analyses of high-dimensional micro-array gene expression data obtained from subjects vaccinated against influenza.