Predicting pesticide dissipation half-life intervals in plants with machine learning models

Area of Science:

• Environmental Chemistry
• Agricultural Science
• Computational Chemistry

Background:

• Pesticide dissipation half-life in plants is vital for environmental risk assessment and setting pre-harvest intervals.
• Empirical measurements of pesticide dissipation are highly variable, making accurate prediction challenging.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting pesticide dissipation half-life intervals in plants.
• To identify key molecular substructures influencing pesticide dissipation rates.

Main Methods:

• Utilized a dataset of 1363 pesticide dissipation half-lives across 311 pesticides, 10 plant types, and 4 plant components.
• Developed four machine learning models: Gradient Boosting Regression Tree (GBRT), Random Forest (RF), Support Vector Classifier (SVC), and Logistic Regression (LR).
• Input features included Extended Connectivity Fingerprints (ECFP), temperature, plant type, and plant component class.

Main Results:

• The GBRT-ECFP model achieved the best performance with an F1-micro_binary score of 0.698 ± 0.010.
• Feature importance analysis highlighted aromatic rings, carbonyl groups, organophosphates, =C-H, and N-containing heterocycles as significant substructures.
• Identified novel dissipation half-life intervals to better account for data variability.

Conclusions:

• Machine learning models, particularly GBRT-ECFP, demonstrate significant utility in predicting pesticide dissipation half-lives in agricultural crops.
• Understanding the role of specific molecular substructures can improve environmental fate assessments of pesticides.
• This approach aids in establishing reliable pre-harvest intervals and ensuring agricultural safety.