Integration of machine learning and pan-genomics expands the biosynthetic landscape of RiPP natural products

摘要: Most clinical drugs are based on microbial natural products, with compound classes including polyketides (PKS), non-ribosomal peptides (NRPS), fluoroquinolones and ribosomally synthesized post-translationally modified (RiPPs). While variants of biosynthetic gene clusters (BGCs) for known products easy to identify in genome sequences, BGCs new escape attention. In particular, evidence is accumulating that RiPPs, subclasses thus far may only represent the tip an iceberg. Here, we present decRiPPter (Data-driven Exploratory Class-independent RiPP TrackER), a mining algorithm aimed at discovery novel classes. DecRiPPter combines Support Vector Machine (SVM) identifies candidate precursors pan-genomic analyses which these encoded within operon-like structures part accessory genus. Subsequently, it prioritizes such regions presence enzymology patterns cluster precursor peptide conservation across species. We then applied mine 1,295 Streptomyces genomes, led identification 42 families could not be found by existing programs. One was studied further elucidated as subfamily lanthipeptides, designated Class V. Two previously unidentified modifying enzymes proposed create hallmark lanthionine bridges. Taken together, our work highlights how product can discovered methods going beyond sequence similarity searches integrate multiple pathway criteria.