A method for joint dimensionality reduction of microbiome and metabolomics data

Access to EGA study: EGAS00001001704 / Lifelines-DEEP

Built upon our previous work on generalized matrix decomposition (GMD), we introduce a new framework for joint dimensionality reduction of structured microbiome data. The latent structures from this new method are informed by the user’s prior knowledge about the relationships between samples and between variables, and therefore are more interpretable. Unlike co-inertia analysis, our framework identifies the joint latent structures shared between data types, which yields systems biology insights into how the gut microbial community works. In practice, however, one may not know a priori whether the auxiliary structures are informative for the target data matrix. To alleviate this concern, we will also provide an adaptive version of GMD that is robust against potential misspecification of prior structures.