Getting poster data...
Cedric C. Laczny, Patrick May, Nikos Vlassis, Paul Wilmes (Luxembourg Centre for Systems Biomedicine, Esch-sur-Alzette, Luxembourg)We propose an approach for the reference-independent visual comparison of microbial community (MC) compositions of metagenomic datasets derived from different conditions, e.g., case vs. control. The approach is based on our recently developed methodology for the alignment-free visualization of metagenomic data by nonlinear dimension reduction of centered log-ratio transformed sequence signatures via Barnes-Hut Stochastic Neighbor Embedding. We introduce here the concept of "condition-specific weights" which allow for an efficient and intuitive identification of condition-specific clusters without prior annotation or taxonomic identification. A proof-of-concept is provided based on in silico simulated MCs. Further, applying our approach to metagenomic data obtained from fecal samples of ten healthy individuals, we reveal condition-specific cluster structures representing microbial populations found specifically in one individual uniquely following a vegetarian diet. Overall, our approach is robust with regards to parameter values, scales well with current dataset sizes, and can be applied even to small cohorts.