Discovering associations in high dimensional imaging-genetics data: A comparison study of dimension reduction and regularisation strategies combined with partial least squares

Brain imaging is increasingly recognised as an intermediate pheno-type in the understanding of the complex path between genetics and behavioural or clinical phenotypes. In this context, a first goal is to propose methods to identify the part of genetic variability that explains some neuroimaging variability. Here, we investigate multi-variate methods, Partial Least Squares (PLS) regression and Canonical Correlation Analysis (CCA), in order to identify a set of Single Nucleotide Polymorphisms (SNPs) covarying with a set of neuroimaging phenotypes derived from functional Magnetic Resonance Imaging (fMRI). Because in such high-dimensional settings multi-variate methods overfit the data, we propose a comparison study of several dimension reduction and regularisation strategies combined with PLS or CCA. We demonstrate that the combination of univariate filtering and sparse PLS outperforms all other strategies and is able to extract a significant link between a set of SNPs and a set of brain regions activated during a reading task.