Computer-assisted prediction of pesticide substructure using mass spectra.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Chemometrics

Background:

• Pesticide residue analysis and lead compound screening are crucial for food safety and drug discovery.
• Accurate identification of pesticide substructures is essential for these processes.
• Existing methods may lack efficiency or accuracy in substructure identification.

Purpose of the Study:

• To develop and validate a computational method for identifying 16 key pesticide substructures using mass spectral data.
• To optimize feature selection for improved accuracy in substructure prediction.
• To enhance the efficiency of pesticide analysis and lead compound screening.

Main Methods:

• Mass spectral data were transformed into 396 features.
• Genetic Algorithm-Partial Least Squares (GA-PLS) was used for feature selection.
• Support Vector Machine (SVM) was employed for validation.
• AdaBoost algorithm combined with Classification and Regression Tree (AdaBoost-CART) was trained for prediction.

Main Results:

• Optimized feature sets were obtained for each of the 16 substructures.
• The optimized feature sets achieved high recognition success rates, mostly between 85% and 100%.
• This approach significantly outperformed using the original 396 features.

Conclusions:

• The developed mass spectral classifiers effectively predict the presence or absence of 16 pesticide substructures.
• The optimized feature selection significantly enhances prediction accuracy.
• This method offers a powerful tool for pesticide residue analysis and screening of pesticide lead compounds.