Metabolic modeling-based tools for integrative microbiome data analysis

Abstract

Complex communities of microbes reside in and on humans, where they closely interact with their hosts by performing a massively diverse array of metabolic reactions. Genomic and metabolomic technologies can now describe both the taxonomic profile of these communities and their metabolic products in unprecedented detail. By measuring both microbial composition and metabolite phenotypes from the same samples, and using the resulting datasets to make and evaluate predictions on the links between microbes and metabolites, it may be possible to infer and characterize metabolic mechanisms occurring in complex natural communities. However, relatively few computational analysis tools exist to integrate and make sense of such “microbiome-metabolome” datasets. In this dissertation, I describe the development and application of methods that use these datasets and reference databases to identify and evaluate relationships between microbes and metabolites. After introducing the current state of knowledge and available tools in the study of how microbial metabolites impact human health and disease, I present an initial framework for integrating microbiome and metabolomics datasets using metabolic modeling. I demonstrate its ability to predict and explain metabolic shifts in bacterial vaginosis, and further illustrate its application in two case studies, deciphering diet-microbiome interactions in mice and characterizing metabolic mechanisms in the microbiota of children with autism spectrum disorder. In order to compare this approach with alternatives and gain a better understanding of the limiting factors in microbiome-metabolome data analysis, I next describe a comprehensive framework for defining gold-standard mechanistic links between microbes and metabolites and using simulations to evaluate and compare our ability to recover them across different datasets and analysis methods. Finally, informed by the previous applications and evaluations, I introduce MIMOSA2, an updated software tool for inferring mechanistic links from microbiome-metabolome datasets. Together, this work reinforces and advances the utility of metabolic modeling for the analysis and interpretation of large-scale microbiome-metabolome studies.