Sparse PLS: Variable Selection when Integrating Omics data

Kim-Anh Lê Cao (1, 2) , Debra Rossow (3) , Christèle Robert-Granié (1) , Philippe Besse (2)

1 SAGA - Station d'Amélioration Génétique des Animaux
2 IMT - Institut de Mathématiques de Toulouse UMR5219
3 Institute for Wine Biotechnology [University of Stellenbosch - Afrique du Sud]

Recent biotechnology advances allow the collection of multiple types of omics data sets, such as transcriptomic, proteomic or metabolomic data to be integrated. The problem of feature selection has been adressed several times in the context of classification, but has to be handled in a specific manner when integrating data. In this study, we focus on the integration of two types of data sets that are measured on the same samples. Our goal is to combine integration and variable selection in a one-step procedure using the good properties of PLS to facilitate the biologists interpretation. A novel computational methodology called sparse PLS is introduced, with two variants depending on the modelling or predictive purpose of the analysis, to deal with these newly arisen problems. The sparsity of our approach is obtained by Lasso penalization of the loading vectors in a SVD PLS version. Sparse PLS is shown to be effective and biologically meaningful. Comparisons with classical PLS are performed on real data sets and a thorough biological interpretation of the results obtained on one data set is provided. We show that sparse PLS not only beneffits from the attractive stability property of PLS but also provide a valuable variable selection tool for high dimensional data sets.

Mots clés: microarray

Domaines: Génomique, Transcriptomique et Protéomique [q-bio.GN]

Kim-Anh Lê Cao, Debra Rossow, Christèle Robert-Granié, Philippe Besse. Sparse PLS: Variable Selection when Integrating Omics data. 2008. ⟨hal-00300204v1⟩