Machine learning-assisted discovery of fungal effectors with antimicrobial activities

Plant-associated fungi, including pathogens as well as mutualists, secrete small proteins, typically referred to as effectors, to support their colonization of host tissues. Although essentially described as modulators of plant immunity, effectors may also function as antimicrobials antagonizing the growth of bacterial and fungal competitors in plant microbiota. Such antimicrobial effectors have recently been identified in the soil-borne pathogen Verticillium dahliae, and their occurrence and conservation throughout the fungal kingdom remain enigmatic. We aimed to annotate genes encoding putative antimicrobial effectors in fungal genomes. Predictors of antimicrobial activity have previously been developed but are mostly dedicated to short peptides and therefore unadapted to fungal effectors. After curating a large set of previously characterized antimicrobial proteins, we trained a classifier that can accurately predict the antimicrobial activity of fungal effectors, relying on sequence- and structure-derived physicochemical properties. This tool was used to predict antimicrobial effector catalogs in fungal genomes, and allows us to screen protein databases to discover novel antimicrobial effector families.