Fig. 2

Microbiome molecular epidemiology. a Multiomic profiling of host and microbiota enables in-depth characterization of community properties from multiple culture-independent data types (including metagenomics, metatranscriptomics, metaproteomics, and metametabolomics) to address questions concerning the microbiome’s composition and function. b As in host-targeted molecular epidemiology, metagenomic and other metaomic data types can be integrated and associated with the available metadata to provide a comprehensive mechanistic understanding of the microbiome. c A wide range of early-stage data analysis choices can strongly affect microbial community data analysis, including the quality control of the raw data, the normalization of the raw data, choice of host and microbial features to extract, and algorithms to profile them. A hypothetical example of four taxonomic features is shown derived from four samples with differing metagenomic sequencing depths (top). Features with the same relative abundances may thus appear to be different on an absolute scale because larger sequencing depth can generate larger read counts (top). Normalization also corrects for potential batch effects and helps to preserve meaningful signal between cases and controls (bottom). Note that the precise methods used for global visualizations, such as the ordination method, can dramatically affect how the data are summarized, as can important parameters in the process, such as the (dis)similarity measures used to compare features or samples. d Within an individual study, the integration of multiple metaomic data types can provide stronger collective support for a hypothesis. Here, a hypothetical disease association is shown at the DNA, RNA, and protein or metabolite levels, providing a more complete picture of the disease pathogenesis. e When they differ between datasets, the strong technical effects that the choices mentioned above have on individual studies can impede multi-study meta-analyses, making this type of population-scale analysis difficult in the microbiome. When possible, the meta-analysis of host and microbial features with respect to shared phenotypes of interest can allow more confidence in prioritizing microbial taxa, gene products, or small molecules that have statistically significant roles in disease relative to covariates. f Finally, as with genome-wide association studies, it is critical to validate putative associations of top candidate microbial features with follow-up experimentation. In the microbiome, this can include studies involving animal models (such as gnotobiotic mice), mammalian cell systems, and/or microbial cultures