Relative count data and its application: assessing the consistency of gut microbial patterns in wild baboons

Abstract

16S rRNA sequencing makes it possible to take compositional snapshots of host-associated microbial communities. The technology has allowed us to learn a great deal about how change in these communities associates with host biology and to generate hypotheses about the roles of gut microbes in host metabolism and immune development. Though this and other applications of next-generation sequencing (NGS) measure compositions (as \textit{relative} feature abundances), these data are often used as proxies for absolute quantities. This dissertation presents two studies: one related to the accuracy of this data and one to its application. In the first project, we used extensive simulated and real data to assess the accuracy of relative (NGS-like) data in a differential abundance testing setting, where compositional effects are expected to introduce spurious differences. We found that the accuracy of differential abundance estimates correlated with key characteristics of the data: larger fold change between experimental conditions and larger proportions of differentially abundant features correlated with increases in the rates of spurious differential abundance calls. Secondly, we used 16S data to investigate the consistency of patterns of variation in gut microbiome composition across hosts. This aspect of gut microbial biology remains understudied partially for want of data. Here we used over 5000 16S profiles from wild baboons collected as a part of the Amboseli Baboon Research Project. We characterized gut microbial correlation patterns within baboon hosts and compared these across hosts, finding a remarkable degree of similarity: pairwise associations between gut bacteria are highly consistent across this population. However, it remains unclear to what extent this consistency of gut bacterial correlation patterns across hosts may be driven by shared diet.