A Galaxy interface to facilitate multi-block analysis

Multi-omics data analysis is one of the main challenges currently faced by integrative biology. Solving it requires combining competences from multiple domains like statistical analysis, computer science and experimental biology. Several types of analytical approaches have been proposed over the past few years in order to perform integrative biology, among them a lot of work around networks (inference or visualization) and more recently around multi-block analyses. The mixOmics package implements multiple statistical analysis methods to integrate different types of omics data (e.g. transcriptomics, metabolomics, proteomics). We focus here on selecting variables in the context of discriminant analysis, where various blocks (each block corresponding to one type of omics) are provided as input.

We built upon the existing mixOmics block.splsda function, which performs feature selection simultaneously on several types of omic data measured on the same individuals, with an emphasis on prediction, and deals with the high number of variables. Sparse PLS-DA is a particular case of SGCCA and therefore exploratory analysis relies on correlation circle plots. We provide additional tools in order to check the possibility of overlaying different correlation circles relative to several blocks. The user can also zoom in on the resulting plot to select subsets of relevant correlated variables. Finally, a network in graphml format is built from selected variables and additional variables of interests. Links are drawn between variables when they are correlated, and the network can be visualized externally using a platform like Cytoscape.

The entire pipeline has been integrated into Galaxy and can be installed from the Toolshed (viscorvar repository). Galaxy XML wrappers and additional R functions source code are also available on GitLab : https://gitlab.com/bilille/galaxy-viscorvar.

Watch a demo (JOBIM 2020 e-poster sessions) : https://meet.jit.si/jobim2020_poster_136

Authors: Maxime Brunin, Pierre Pericard, Guillemette Marot - June 2020