Chapter 5 Selection of interaction effects in compressed multiple omics representation

This chapter is organized as follows. Section 5.2 introduces the setting related to linear models of interactions and proposes a framework to learn with complementary datasets. Section 5.3 describes the method, entitled SICOMORE for Selection of Interaction effects in COmpressed Multiple Omics REpresentation, which combines hierarchical clustering, data compression, variable selection with a lasso procedure and model testing for recovering relevant interactions. Our approach is illustrated with numerical simulations in Section 5.4 and with an application to study the interactions between the genome of the species Medicago truncatula and the microbial community of its rhizosphere in Section 5.5.