|
Abstract:
|
Human microbiome studies aim to characterize the microbial communities in the body and the effect of environmental factors on them. A range of experimental techniques have been developed in recent years to catalogue the species composition of a biological sample through ribosomal DNA sequencing, to measure the transcription level of microbial genes, and the synthesis of proteins and metabolites. Modelling such heterogeneous data with a coherent assessment of uncertainty from exploratory analysis, through model selection and inference presents a significant challenge. We propose a Bayesian approach based on latent factors, which is capable of combining insights from various experiments in a parsimonious and interpretable way. We discuss how to scale up computations to massive datasets and evaluate the robustness to prior parameters.
|
