Microbiome-driven machine learning for predicting suppressiveness to Rhizoctonia solani in organic-amended soils

Abstract

This study introduces a microbiome-integrated machine learning framework to predict soil suppressiveness against Rhizoctonia solani, a destructive fungal pathogen in crops. Combining bacterial and fungal community data, edaphic factors, and disease severity metrics, this study developed robust classification and regression models using Artificial Neural Networks, Random Forest, and Support Vector Machines. Random Forest emerged as the top-performing model, achieving the highest accuracy and explanatory power in classification (98.56 %) and regression (95.40 %) tasks. The study also introduced the Biocontrol Index (BCI), which integrates microbial abundance and predictive importance into interpretable indices linked to soil suppressiveness. Additionally, ecological indices, including alpha diversity, evenness, and beta diversity, were calculated per sample to provide insights into microbial community structure and dynamics. Key predictive features included soil amendments with high microbial activity, carbon-to‑nitrogen ratios, and nutrient levels like calcium and manganese. This integration of microbiome data with machine learning offers a novel, data-driven approach to tackle the high dimensionality of amplicon-sequencing data while also understanding and predicting soil suppressiveness.