Identifying microbiota community patterns important for plant protection using synthetic communities and machine learning

Abstract

Plant-associated microbiomes contribute to important ecosystem functions such as host resistance to biotic and abiotic stresses. The factors that determine such community outcomes are inherently difficult to identify under complex environmental conditions. In this study, we present an experimental and analytical approach to explore microbiota properties relevant for a microbiota-conferred host phenotype, here plant protection, in a reductionist system. We screened 136 randomly assembled synthetic communities (SynComs) of five bacterial strains each, followed by classification and regression analyses as well as empirical validation to test potential explanatory factors of community structure and composition, including evenness, total commensal colonization, phylogenetic diversity, and strain identity. We find strain identity to be the most important predictor of pathogen reduction, with machine learning algorithms improving performances compared to random classifications (94-100% versus 32% recall) and non-modelled predictions (0.79-1.06 versus 1.5 RMSE). Further experimental validation confirms three strains as the main drivers of pathogen reduction and two additional strains that confer protection in combination. Beyond the specific application presented in our study, we provide a framework that can be adapted to help determine features relevant for microbiota function in other biological systems.

Abstract

The authors investigate microbiota properties for plant protection using synthetic communities and machine learning approaches. They identify strains that reduce pathogen colonization despite variation in microbiota composition.