Computationally Mining the Microbiome for Biologically Meaningful Results

Abstract

The human microbiome—the diverse commensal microbial communities that reside in and on every individual—has received considerable attention over the past few decades for its vital role in modulating host physiology. Although recent years have witnessed an explosion of meta-omics data generated from high-throughput sequencing-based microbiome studies, our knowledge of how specific commensal bacteria regulate human health and disease remains disproportionately limited. Developing computational frameworks that allow one to coherently derive biologically meaningful results given a collection of distinct ‘omics datasets will contribute to an improved understanding of the complex, disease-modulatory host–microbiome interplay. In this dissertation, I established several bioinformatic pipelines that help accomplish this goal and applied them to investigate the role of the microbiome in different aspects of viral infectious diseases. Specifically, in Chapter 2, I demonstrated an analysis scheme that links phenotype-associated taxa to the metabolic pathways they significantly contribute to and correlated multiple high-dimensional datasets obtained from microbiome analysis, metabolomics, and immune phenotyping. Applying this workflow, I found that a commensal Sutterella species, its metabolic pathways, and the production of short-chain fatty acids and bile acids are associated with increased antibody responses to an HIV vaccine. In contrast, in Chapter 4, I provided an alternative procedure that connects differentially abundant metabolic pathways with their main contributing reactions, genes, and microbes and correlated the microbiome functional profile with the cytokine milieu quantified by multiplex immunoassays. Applying this workflow, I found that symptomatic COVID-19 infection results in a less diverse gut microbiota marked by lower functional capacity for the tyrosine biosynthesis pathway, which—in addition to being associated with interferon alpha 2a levels—is driven by a reduction in prephenate dehydrogenase that can be primarily attributed to just a few bacterial species. Moreover, in Chapter 3, I extended a computational discovery platform our group has previously developed to create highly-specific data-driven hypotheses that we further validated experimentally using in vitro assays and bacterial genetics. I also established a straightforward bioinformatic pipeline that efficiently extracts the gene homologs of interest from expansive abundance tables produced by metagenomic functional analysis, enabling us to quickly confirm the translational relevance of our findings in human-derived datasets. Ultimately, we delineated a complete microbiota-driven pathway—including identification of the specific bacterial taxa, bacterial gene, bacterial metabolite, and host receptor involved—that broadly inhibits viral infections. Last but not least, in Chapter 5, I presented an innovative method inspired by concepts from topological data analysis to compute the within-sample phylogenetic microbial biodiversity. Using data from earlier chapters for benchmarking, I showed that this new approach efficiently and accurately estimates Faith’s phylogenetic diversity compared to the standard path. Taken together, this dissertation offers several generalizable computational frameworks that assist the analysis of heterogeneous meta-omics datasets in a focused, directed, and interpretable fashion, with applications that have remarkably enhanced our knowledge of the microbiome in various facets of viral infectious diseases, including vaccine response, viral acquisition, and disease severity.