Integrating microbiome and machine learning for precision diagnosis of rice bakanae disease

Bakanae is a fungal rice disease that is threatening global rice production, causing severe yield losses. The plant microbiome plays a significant role in plant stress resistance, but its high-dimensional characteristics have not been fully exploited. Therefore, we integrated the microbiome and machine learning (ML) to diagnose bakanae disease in this study. We found significant correlations between Gammaproteobacteria and Bacteroidia and the severity of bakanae disease. We constructed different diagnosis models based on random forests (RF), support vector machines (SVM), and convolutional neural networks (CNN) on 88 biological replicates with an independent test set. We found that the RF model demonstrated strong performance across four taxonomic levels, with an accuracy of 88.9% and an F1 score of 94.1%. Notably, a Bray-Curtis dissimilarity-based extraction method was proposed to rapidly screen practical information from the original microbial community, which can enhance the model performance to a certain extent. According to phenotypic data, the disease severity of infected samples was classified into two levels (high and low infected levels) using the K-means clustering method. In the diagnosis of infection severity based on the family level, the model's prediction accuracy reached 77.8%. Collectively, these findings highlight that the combination of microbiome with ML can advance diagnostic strategies for bakanae disease, providing new avenues for precision agriculture.