Combined interaction of fungicides binary mixtures: experimental study and machine learning-driven QSAR modeling

Area of Science:

• Agricultural Science
• Computational Chemistry
• Plant Pathology

Background:

• Fungicide resistance is a major threat to crop production.
• Mixture strategies are crucial for managing fungicide resistance.
• Predictive modeling can optimize the development of effective fungicide mixtures.

Purpose of the Study:

• To evaluate the fungicidal activity and interactions of binary fungicide mixtures.
• To develop and compare Quantitative Structure-Activity Relationship (QSAR) models for predicting fungicidal efficacy.
• To identify optimal fungicide combinations for delaying fungicide resistance.

Main Methods:

• Generated fifty binary fungicide mixtures using a fixed ratio ray design.
• Analyzed mixture interactions using Combination Analysis (CA) and Interference Analysis (IA) models.
• Employed Multiple Linear Regression (MLR), Support Vector Machine (SVM), and Artificial Neural Network (ANN) for QSAR modeling.

Main Results:

• Most fungicide mixtures exhibited additive interactions.
• The CA model was more accurate than the IA model for predicting fungicidal activity.
• Machine learning (ML)-based models (ANN and SVM) outperformed MLR in predictive performance.
• The ANN model demonstrated superior predictability (R²=0.91, R²cv=0.81, R²test=0.845) compared to SVM (R²=0.91, R²cv=0.78, R²test=0.77).

Conclusions:

• ML-based QSAR models are valuable tools for designing effective fungicidal mixtures.
• The developed ANN model shows high potential for predicting fungicidal activity and guiding resistance management strategies.
• Optimizing fungicide combinations is key to sustainable agriculture and mitigating resistance development.