A key challenge in the analysis of longitudinal microbiomes data is to go beyond computing their compositional profiles and infer the complex web of interactions between the various microbial taxa, their genes, and the metabolites they consume and produce. To address this challenge, we developed a computational pipeline that first aligns multi-omics data and then uses dynamic Bayesian networks (DBNs) to integrate them into a unified model. We discuss how our approach handles the different sampling and progression rates between individuals, how we reduce the large number of different entities and parameters in the DBNs, and the construction and use of a validation set to model edges. Applying our method to data collected from Inflammatory Bowel Disease (IBD) patients, we show that it can accurately identify known and novel interactions between various entities and can improve on current methods for learning such interactions. Experimental validations support several predictions about novel metabolite-taxa interactions. The source code is freely available under the MIT Open Source license agreement and can be downloaded from https://github.com/DaniRuizPerez/longitudinal\_multiomic\_analysis_public.

Dynamic Bayesian networks for integrating multi-omics time-series microbiome data