Nonlinear SAR Modelling of Mosquito Repellents for Skin Application.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Toxicology

Background:

• Discovering novel mosquito repellents is crucial for public health, particularly against disease vectors like Aedes aegypti.
• Traditional repellent development is lengthy and resource-intensive, necessitating innovative approaches.
• Computational modeling offers a promising avenue to optimize repellent discovery processes.

Purpose of the Study:

• To develop and validate a predictive Structure-Activity Relationship (SAR) model for identifying potential Aedes aegypti mosquito repellents.
• To leverage artificial neural networks (ANNs) for enhanced accuracy in predicting molecular repellent activity.
• To expedite the identification of novel repellent candidates through computational screening.

Main Methods:

• A dataset of 2171 molecules with known Aedes aegypti repellent activity was compiled.
• Information-rich molecular descriptors were used as input for a three-layer perceptron (TLP) artificial neural network.
• A 20/6/2 TLP architecture demonstrated high accuracy on training and test datasets, with 94% and 89% respectively.

Main Results:

• The optimized 20/6/2 TLP model achieved high predictive accuracy for mosquito repellent activity.
• Cross-validation using all molecules as both training and test members confirmed model robustness.
• Analysis of prediction errors provided insights into model limitations and chemical space.
• The model successfully predicted the activity of novel molecules, two of which were validated in vivo.

Conclusions:

• The developed SAR model, based on ANNs, significantly enhances the efficiency of mosquito repellent discovery.
• This computational approach accelerates the identification of promising repellent candidates for further development.
• The validated model holds potential for screening large chemical libraries to find effective Aedes aegypti repellents.