Genomics- and Machine Learning-Accelerated Discovery of Biocontrol Bacteria

Abstract

Microorganisms with antimicrobial activity have been used to successfully control various plant pathogens. The discovery of organisms with protective activity depends on empirical screenings to assess microbial activity against pathogens of interest. Machine learning can accelerate the discovery process by making screening and, thus, discovery more efficient. We developed a novel machine-learning workflow to identify genomic features associated with fungicidal activity of bacteria, and leveraged those genomic features to discover additional bacteria with the desired activity. We applied our workflow to discover solutions to two problematic fungal diseases: sorghum anthracnose and black sigatoka of banana. These diseases are problematic worldwide, with a particularly devastating impact on small-holder farmers in Sub-Saharan Africa. We screened a total of 1,227 bacterial isolates for antifungal activity against these pathogens using detached-leaf methods and identified 72 taxonomically diverse isolates with robust activity against one or both of these pathogens. We identified biosynthetic gene clusters associated with activity against each pathogen. Machine learning improved the discovery rate of our screen by threefold, and led to the discovery of a taxonomic group in which fungicidal activity has never been reported. This work highlights the wealth of biocontrol mechanisms available in the microbial world for management of fungal pathogens, generates opportunities for future characterization of novel fungicidal mechanisms, and provides a set of genomic features and models for discovering additional bacterial isolates with activity against these two pathogens. Finally, our workflow generalizes to any discovery effort where genomic information is available to guide candidate selection.[Formula: see text] Copyright © 2021 The Author(s). This is an open access article distributed under the CC BY 4.0 International license .