From metabolic reprogramming to diagnostic tool: A machine learning framework identifies biomarkers for soil microplastic stress in rice.

As emerging environmental contaminants, microplastics (MPs) increasingly infiltrate agricultural ecosystems, threatening crop safety and productivity. While the phytotoxic effects of MPs on plants like rice are becoming apparent, a significant gap exists in developing high-throughput tools for diagnosing MP contamination levels in soil. This study integrates untargeted metabolomics with machine learning to comprehensively elucidate the metabolic responses of rice to MP exposure and to establish a diagnostic model. Our results demonstrate that MPs significantly disrupt key metabolic pathways, including ABC transporters, the tricarboxylic acid (TCA) cycle, and alanine, aspartate, and glutamate metabolism. A machine learning-based feature selection pipeline identified a minimal panel of seven metabolic biomarkers. A predictive model built on this panel accurately distinguished rice samples under different soil pollution levels (Low, Medium, High) and demonstrated robust diagnostic capability in an independent external validation set (ROC-AUC = 0.90). This integrative analytical framework provides a scalable strategy for monitoring and early warning of microplastic pollution in agricultural systems, offering significant implications for ensuring food security and technical support for the future promotion of in-situ soil microplastic risk assessment methods.