Background:

Understanding the complex relationships between bacteria and metabolites in ecological systems are extremely important in studies of different microbiomes. Longitudinal multi-omics study is adopted to investigate interactions between bacteria and metabolites, by directly associating their longitudinal profiles. Since a bacteria/metabolite may involve in many different biological processes, the longitudinal profile is an average of different interactions. Therefore, direct association could only uncover the strongest interactions. Results Here we present a computational approach that can rebuild short- and long-term bacteria-metabolite interactions from longitudinal multi-omics datasets. For this task, we re-analyse data (both microbial sequencing and metabolomic analysis) from an in vitro model of Clostridioides difficile infection and faecal microbiota transplant, a disease state and mode of therapy in which perturbed microbiome-metabolome interactions (and their reversal) are well-established to be pertinent. By analysing such a dataset, we generated both a short-term and a long-term interaction network, which predicted many new interactions. Four new interactions were randomly selected to be validated. In batch culture experiments, we validated two of them: (1) Ruminococcus gnavus and Ruminococcus luti could generate 3-ketocholanic acid (2) Blautia obeum could consume succinate. Conclusions The deconvolution of the raw longitudinal signal into short- and long-term trends can help users to gain a deeper understanding of their data. This tool will be useful for high-throughput screening of microbe/metabolite/host interactions from a longitudinal multi-omics setting.