Model for prediction of pesticide residues in soybean oil using partial least squares regression with molecular descriptors selected by a competitive adaptive reweighted sampling algorithm

We developed a partial least squares regression (PLSR) model based on competitive adaptive reweighted sampling (CARS) to predict the processing factors of 54 pesticides during soybean oil processing. Characteristic variables were selected to improve the performance of the model. Four calculators were used to compute the molecular descriptors used in the model, and the model based on values computed with ChemoPy produced the best results: Rc ​= ​0.94, RMSEc ​= ​0.67, Rp ​= ​0.91, and RMSEp ​= ​0.54 for hot-pressed oil and Rc ​= ​0.93, RMSEc ​= ​0.73, Rp ​= ​0.93, and RMSEp ​= ​0.59 for cold-pressed oil. A rapid and quantitative model of processing factors was established to predict the behaviour and distribution of pesticide residues during food processing. The model was further validated using data from field-grown soybeans; it demonstrated a high correlation coefficient between predicted and measured residue concentrations (Rp ​> ​0.93, RMSEp ​< ​0.72) and successfully predicted the distribution and behaviour of pesticide residues. Our model provides a reference for assessing safety risk and determining the maximum residue limits for pesticides in processed products.