The effects of pre-processing techniques on network analysis of microbiome sequencing data

Abstract

The human microbiome is a growing area of research. Enabled by advances in sequencing techniques, vast amounts of microbiome data are being generated. Using this data to answer research questions is challenging due to the compositionality, sparsity and high-dimensionality of the data. Network techniques have been succesfully applied to make sense of microbiome data, yet difficulties still remain. Here, we compare different methods of preparing data for the use in weighted gene coexpression network analysis (WGCNA), a popular framework within bioinformatics utilizing network theory. Three different methods were applied: one based on simple compositionality, one based on the centered log-ratio transform, and one using SparCC: an advanced method specifically designed to infer correlations in microbiome sequencing data. We found that network results vary widely depending on the method used.