Three-Level cascading defense mechanism of silicon-based materials for mitigating heavy metals stress in crops.

Silicon-based materials (SBMs) are widely used to alleviate heavy metals (HMs) stress in crops, yet their integrated effects on soil-plants systems remain insufficiently understood owing to the complex interactions among soil properties, amendment characteristics and experimental conditions. This study presents a global meta-analysis to comprehensively evaluate the effects of SBMs on crop response to HMs stresses, and machine learning (ML) approaches are employed complementarily to identify key drivers and predict HMs accumulation in grain (GHMs) following SBMs application. Our findings reveal that SBMs application increased soil available silicon by 42.40%, reduced HMs transfer from roots to grains by 19.25%, and enhanced photosynthetic performance by 44.02%, while decreasing GHMs by an average of 29.90%. Together, these results indicate that SBMs alleviate HMs stress through coordinated effects on soil immobilization, internal HMs transfer, and plant physiological resilience, supporting a three-level cascading soil-plant defense mechanism. Among 13 ML models, the AdaBoost algorithm delivered the optimal predictive performance with a coefficient of determination (R2) ranging from 0.84 to 0.95. SHAP analysis identified soil HMs content as the most important predictor of GHMs, followed by soil available P, fertilizer P application rate, and soil pH. The predicted median grain HMs concentration (0.255 mg/kg) closely matched the observed value (0.276 mg/kg). By integrating mechanistic understanding with interpretable prediction, this study provides a framework for optimizing SBMs selection and application strategies based on site-specific soil conditions, which has important implications for the risk control and sustainable management of contaminated farmlands.