Exploring Insecticidal Molecules with Random Forest: Toward High Insecticidal Activity and Low Bee Toxicity.

Area of Science:

• Agricultural Science
• Computational Chemistry
• Toxicology

Background:

• High-potency insecticidal molecules are vital for crop protection and global food security.
• However, widespread use is hindered by significant toxicity to non-target organisms, particularly bees.
• Developing pesticides that balance efficacy with environmental safety is a critical agricultural challenge.

Purpose of the Study:

• To introduce a novel strategy for identifying insecticidal molecules with both high pest efficacy and low bee toxicity.
• To leverage machine learning (ML) for predicting these dual properties.
• To validate the predictive models through experimental testing.

Main Methods:

• Synthesis and testing of novel insecticidal molecules to train a ML model for pest activity prediction.
• Training a separate ML model using public data to predict bee toxicity.
• Integration of both ML models to predict and identify candidate molecules.
• Experimental validation of a predicted molecule exhibiting desired properties.

Main Results:

• Machine learning models achieved high predictive performance: mean AUC of 0.88 ± 0.05 for insecticidal activity and 0.91 ± 0.01 for bee toxicity.
• The integrated dual-ML-model approach successfully identified a molecule with high insecticidal activity and low bee toxicity.
• Experimental validation confirmed the predicted efficacy and safety profile of the identified molecule.

Conclusions:

• The dual-ML-model approach provides a powerful and efficient strategy for discovering novel insecticidal compounds.
• This method facilitates the development of effective and environmentally benign pesticides, supporting sustainable agricultural practices.
• This research paves the way for safer pest management solutions, balancing crop protection with pollinator health.