Predicting the aquatic toxicity mode of action using logistic regression and linear discriminant analysis

Area of Science:

• Environmental Chemistry
• Toxicology
• Computational Chemistry

Background:

• Accurate classification of chemical aquatic toxicity is crucial for environmental risk assessment.
• Predictive models are needed to understand chemical toxic modes of action.
• Existing methods may require improvement for robustness and interpretability.

Purpose of the Study:

• To develop and validate statistically significant, robust, and predictive models for classifying chemical aquatic toxic modes of action.
• To compare the performance of logistic regression (LR) against linear discriminant analysis (LDA).
• To explore the relationship between chemical structure descriptors and aquatic toxicity.

Main Methods:

• Logistic regression (LR) was employed for model construction.
• Stepwise forward discriminant analysis (LDA) was used for predictor selection from experimental data and structure-based descriptors (CODESSA, DRAGON).
• Internal and external validation, leverage approach for applicability domain, and influence plots for diagnostics were utilized.

Main Results:

• Logistic regression models demonstrated statistical significance, robustness, and predictive ability.
• LR generally outperformed LDA, particularly for predicting highly toxic compounds.
• Models were simple, interpretable, and validated for reliability.

Conclusions:

• Logistic regression is a suitable and attractive method for classifying chemical aquatic toxic modes of action.
• The developed models provide reliable predictions and insights into structure-activity relationships.
• This approach enhances the assessment of environmental risks associated with chemical substances.