Causal inference on microbiome-metabolome relations via in silico in vivo association pattern analyses

Abstract

The effects of the microbiome on the hosts metabolism are core to understanding the role of the microbiome in health and disease. Herein, we develop the paradigm of in silico in vivo association pattern analyses, entailing a methodology to combine microbiome metabolome association studies with in silico constraint-based microbial community modelling. By dissecting confounding and causal paths, we show that in silico in vivo association pattern analyses allows for causal inference on microbiome-metabolome relations in observational data. Then, we demonstrate the feasibility and validity of our approach on a published multi-omics dataset (n=346), demonstrating causal microbiome-metabolite relations for 43 out of 53 metabolites from faeces. Finally, we utilise the identified in silico in vivo association pattern to estimate the microbial component of the faecal metabolome, revealing that the retrieved metabolite prediction scores correlate with the measured metabolite concentrations, and they also reflect the multivariate structure of the faecal metabolome. Concluding, we integrate with hypothesis free screening association studies and knowledge-based in silico modelling two major paradigms of systems biology, generating a promising new paradigm for causal inference in metabolic host-microbe interactions.