2026 Theses Doctoral

Computational methods for the analysis of variable genomic content in microbiomes

Coleman, Izaak

Our understanding of the organization of all living things has been tremendously deepened by studying microbes alone. By observing how microbes interact, the discovery of compounds that can halt fatal infections in their tracks have provided relief for millions, while better control over how microbes contribute to agriculture and food production has provided sustenance for billions. Microbes typically exist in complex communities called microbiomes, and recent research into these communities has underscored the highly variable and dynamic nature of microbial genomes.

A core objective of microbiome research is to understand how this variability impacts human health. Researchers can now gain access to the genetic content of entire microbiomes through metagenomics, which measures all genomes in the community as an environmental mixture, and which provides a means to compare genetic variation. It therefore has immense potential to help researchers understand the genetic and mechanistic underpinnings of host-microbiome associations. Due to the complexity of metagenomic data, analysis of genomic variability in microbiomes requires computational methods.

The central problem this thesis addresses is how to represent this variability in the most effective way for comparative microbiome analysis. We introduce new computational methods to: represent and compare variability across microbiomes, evaluate computational pipelines that recover individual genomes from metagenomics data, and to probabilistically model the variable genomic content across multiple microbiome datasets, which improves genome recovery.

These methods blend the high-throughput nature of metagenomic analysis with the high-resolution of comparative genomics. Vancomycin-resistant enterococcus (VRE), a highly resistant, and opportunistic ESKAPE pathogen is a substantial burden on healthcare. Using a gut metagenomic dataset from liver transplant patients, we demonstrate the analytical advantage our methods provide by identifying variable elements of the VRE genome that are predictive of colonization persistence. Furthermore, with our methods we identify key performance gaps in state-of-the-art approaches for recovering individual genomes from metagenomes, and we demonstrate that our probabilistic approach to genome recovery can substantially reduce these gaps.