Predictive and comparative network analysis of the gut microbiota in type 2 diabetes

Metagenome-wide analysis studies provide a unique set of microbial features for biomarker discovery of associated disease as well as for studying diversity and dynamics of microbial communities under different conditions. Taxonomic classification of microbes in metagenomic samples quantifies what microbes are present and in what proportion. Despite the availably of several computational taxonomy profiling methods, this crucial step in metagenome-wide analysis remains very challenging and how using different taxonomy profiling methods might influence the outcome of the analysis is not wellstudied. In this work, we consider three taxonomy profiling methods (MetaPhlAn2, Kraken, and EBI metagenomics pipeline) and examine their effect on the outcome of metagenome-wide analysis based on machine learning and comparative network approaches. Our results suggest that Kraken OTU-based data representation yields the best performing classifiers even using less number of features (e.g., OTUs). In addition, our preliminary results underscore the viability of leveraging multiple taxonomic classification methods in microbial network analysis. Finally, our analysis results are consistent with the current knowledge and reveal novel insights into interaction relationships between potential biomarkers in the gut microbiome associated with T2D.