Improved bacteria population structure analysis on thousands of genomes using unsupervised methods

Over ten thousand genomes of Escherichia coli are now available, and this number will continue to grow for this and other important microbial species. The first approach often used to better understand microbes is phylogenetic group analysis followed by pan-genome analysis of highly related genomes. Here, we combine sequence-based features with unsupervised clustering on up to 2,231 E. coli genomes and a total of 1,367 Clostridium difficile genomes. We show that Non-negative Matrix Factorization (NMF) can identify “mixed”/cryptic genomes, and can better determine inter-related genome groups and their distinguishing features (genes) relative to prior methods.