Support vector machine-based model for toxicity of organic compounds against fish

Area of Science:

• Environmental Chemistry
• Ecotoxicology
• Computational Chemistry

Background:

• Assessing chemical toxicity to fish is vital for environmental protection.
• Quantitative Structure-Activity Relationship (QSAR) models offer a predictive approach to ecotoxicology.

Purpose of the Study:

• To develop a robust quantitative structure-toxicity relationship (QSTR) model for predicting the toxicity of organic chemicals to fish.
• To utilize a chemometric approach combining Support Vector Machine (SVM) and genetic algorithms for enhanced predictive accuracy.

Main Methods:

• A QSTR model was developed using Support Vector Machine (SVM) and genetic algorithms.
• The model was trained on 840 organic compounds and validated on a separate set of 281 compounds.
• Only six molecular descriptors were employed to build the predictive model.

Main Results:

• The developed QSTR model achieved determination coefficients (R²) above 0.70 for both training and test datasets.
• The model demonstrated superior statistical performance compared to existing QSTR models in the literature.
• The study successfully applied SVM and genetic algorithms for predicting fish toxicity (pLC50).

Conclusions:

• The developed SVM-based QSTR model is effective for predicting the toxicity of organic chemicals to fish.
• This approach offers a computationally efficient method for ecotoxicological assessment, requiring fewer molecular descriptors.
• The findings support the use of chemometric methods in environmental risk assessment.