UCSF

High-throughput sequencing has firmly established itself as the leading method for assaying the structure and functional capacity of microbial communities. With this deluge of data, care must be taken to account for technical and biological artifacts in order to produce robust candidate biomarkers. Of particular interest is the use of mixed effects models and nonlinear models to assess key differences between healthy and diseased individuals that arise over time. In my thesis work, I analyzed data from a longitudinal study of inflammatory bowel disease in mice with the aim of uncovering biological features predictive of abnormal microbiome development in the context of chronic inflammation. My analysis uncovered multiple taxa and gene families that have differential temporal trajectories, as well as a few gene families that stratify the diseased and wild type subjects early on. This investigation led to a follow-up study of the underrepresented microbial genomes present in lab mice, to expand our knowledge of the model animal’s microbiome. Since the majority of microbiome studies aimed at future clinical impact are carried out in mice, it is important to know what separates human microbiomes from those of mice, in order to limit hypotheses that are not transferrable. We found that even a modest single cell sequencing effort leads to an appreciable gain in phylogenetic diversity and significantly improves the recruitment of short reads from unrelated mouse metagenomes. Overall, I have shown that robust findings are possible even with a limited set of subjects if one leverages a nuanced statistical modeling approach and undertakes targeted acquisition of new data.