Our lab specialises computational methods and software developments, as well as the application of our methods and tools to biological data sets generated by our collaborators.
Data integration methods using multivariate projection-based methodologies
Our dimension reduction methods are based on the Projection to Latent Structures algorithm (PLS, a term we prefer to Partial Least Squares regression, Wold et al. 2001) that are combined with LASSO regularization to identify important biological features or biomarkers in large-scale biological data sets. Our latest frameworks include DIABLO (Singh et al. 2019) to integration multiple data sets measured on the same N samples (N-integration); MINT (Rohart et al. 2017a) to integrate independent studies measured on the same P variables / genes (P-integration) and mixMC (Lê Cao et al. 2016) for the multivariate analysis of microbial communities, timeOmics (Bodein et al. 2020) to integrate microbiome and ‘omics time course data.
We are interested in developing new multivariate methodologies to
- integrate multi-omics single cell data on matching cells (multiome ATAC-seq, CITE-seq, scNMT-seq)
- integration of multi-omics time course data
- integrate multi-omics and microbiome data
mixOmics R toolkit package (www.mixOmics.org)
mixOmics is one of the few R package dedicated to the integration of multiple ‘omics data (19 novel methodologies implemented so far, amongst which 13 were developed by our lab) and with an increasing uptake from the research community. The package is in the top 5% downloads in Bioconductor ( ~ 80K downloads / year). Check our our recent publication (Rohart et al. 2017b) and and 50-min webinar overview about this project. The mixOmics team run multiple day workshops for an introduction to multivariate projection-based methods for data integration using mixOmics, see our website www.mixOmics.org for news, tutorials, online courses. We continuously improve our software to support our growing userbase.
Multivariate methods for microbiome studies
There are major statistical and computational challenges in analysing microbial communities that currently hinder the potential of microbiome research to substantially advance biomedical understanding. We are currently expanding mixMC to better characterise and understand important microbiome-host interactions. Some of our methods developments aim at addressing batch effects in microbiome experiments and analyse scarce temporal sampling in time course studies.
We analyse microbiome datasets from our collaborators for a wide range of studies, including investigating the role of gut and oral microbiome in spondyloarthropathy diseases, the development of intestinal or salivary microbiota in toddlers and infants, investigating the gut-brain crosstalk in Huntington’s disease, anaerobic digestion in fermenters.