Heuristic and Machine Learning Based Methods for High Throughput Lasso Peptide Discovery

Abstract

This paper outlines a targeted, high throughput approach to discovering lasso peptides, which belong to a class of ribosomally synthesized and post-translationally modified peptides (RiPPs). The method employed for this genome mining effort combines a expectation maximization (EM) search for a lasso peptide biosynthetic gene cluster with a pattern based search for lasso peptide sequences themselves, allowing the software to identify a set of possible candidates from genomes rapidly in time approximately linear to the length of a genome. Two filtering algorithms are employed in series, with a fast “primary” ranking method to discern candidates, and a selective “secondary” ranking scheme to select from this set the most likely real lasso peptides. The results are then presented using a custom front end solution that allows easy visual inspection of biosynthetic gene clusters, the proposed lasso peptide candidates, and their corresponding locations in NCBI sequenced genomes when available. Finally, future directions are discussed as further empirical validation of lasso peptide candidates can be used to augment the predictive power of this tool.