A novel method for predicting bioconcentration factor in rice based on the quantitative ion character-activity...

Area of Science:

• Environmental Science
• Agricultural Science
• Computational Chemistry

Background:

• Heavy metals (HMs) in agricultural soils are a major concern for rice safety.
• Traditional bioconcentration factor (BCF) measurements are costly and time-consuming.
• Quantitative Structure-Activity Relationship (QSAR) is established for organic pollutants, but Quantitative Ion Character-Activity Relationship (QICAR) for inorganic pollutants in plants is less explored.

Purpose of the Study:

• To develop and validate a QICAR model for predicting heavy metal bioconcentration factors (BCF) in rice.
• To assess the efficacy of machine learning algorithms in QICAR modeling for inorganic pollutants.
• To identify key soil and metal properties influencing heavy metal accumulation in rice.

Main Methods:

• Utilized a dataset of 529 soil-rice samples.
• Employed three machine learning algorithms: Random Forest (RF), CatBoost (CAT), and XGBoost (XGB).
• Input features included physicochemical properties of soil and metals.

Main Results:

• Developed highly accurate QICAR models for predicting heavy metal BCF in rice.
• The best CatBoost model achieved R² = 0.91 and MAE = 0.1916.
• Identified the softness index (Σp) as a critical factor with a threshold effect on HM enrichment.

Conclusions:

• QICAR models, particularly those using CatBoost, can accurately predict heavy metal BCF in rice.
• Minimal input features (1-2 metal properties + soil properties) are sufficient for high accuracy.
• The model provides a scientific basis for risk assessment and decision-making regarding agricultural product safety, as demonstrated in Hainan Island.