Composition identification and functional verification of bacterial community in disease-suppressive soils by machine learning.

It has been widely reported that probiotic consortia in the rhizosphere can enhance the plant resistance to pathogens. However, the general composition and functional profiles of bacterial community in soils which suppress multiple diseases for various plants remain largely unknown. Here, we combined metadata analysis with machine learning to identify the general patterns of bacterial-community composition in disease-suppressive soils. Disease-suppressive soils significantly enriched Firmicutes and Actinobacteria but showed a decrease in Proteobacteria and Bacteroidetes. Our machine-learning models accurately identified the disease-conducive and -suppressive soils with 54 biomarker genera, 28 of which were potentially beneficial. We further carried out a successive passaging experiment with the susceptible rps2 mutant of Arabidopsis thaliana invaded by Pseudomonas syringae pv. tomato DC3000 (avrRpt2) for functional verification of potential beneficial bacteria. The disease-suppressive ability of Kribbella, Nocardioides and Bacillus was confirmed, and they positively activated the pathogen-associated molecular patterns-triggered immunity pathway. Results also showed that chemical control by pesticides in agricultural production decreased the disease-suppressive ability of soil. This study provides a method for accurately predicting the occurrence of multiple diseases in soil and identified potential beneficial bacteria to guide a wide range of multiple-strain biological control strategies in agricultural management.