Introducing group-sparsity and orthogonality constraints in RGCCA

RGCCA-a fast and flexible method-generalizes many other well-known methods in order to analyze data-sets comprising multiple blocks of variables. Here we extend RGCCA by adding two new constraints to the RCCCA optimization problem: 1) group sparsity and 2) orthogonality of the block weight vectors. These two constraints facilitate the interpretability of the results when analyzing high dimensional data with a group structure. We illustrate this new method-called gSGCCA-with the analysis of pediatric high-grade glioma data: a set comprising three data blocks. This analysis shows that these new constraints greatly improve the interpretability of the statistical analysis.