Predicting the genotoxicity of polycyclic aromatic compounds from molecular structure with different classifiers.

Area of Science:

• Computational toxicology
• cheminformatics
• QSAR modeling

Background:

• Genotoxicity assessment of polycyclic aromatic compounds (PACs) is vital for public health and environmental safety.
• Traditional genotoxicity testing can be time-consuming and resource-intensive.
• Predictive models using molecular structure can streamline the assessment process.

Purpose of the Study:

• To develop and evaluate computational models for predicting the genotoxicity of 277 PACs based on their molecular structures.
• To compare the performance of k-nearest neighbor (k-NN), linear discriminant analysis, and probabilistic neural network classifiers.
• To establish a consensus model for improved genotoxicity prediction.

Main Methods:

• Calculation of numerical descriptors (topological, geometric, electronic, polar surface area) from PAC molecular structures.
• Measurement of genotoxicity using the SOS Chromotest, quantified by the maximal SOS induction factor (IMAX).
• Development and validation of binary classification models (k-NN, LDA, PNN) and a consensus model.

Main Results:

• The k-nearest neighbor (k-NN) model demonstrated the highest predictive ability with a 93.5% training set classification rate.
• A consensus model, integrating k-NN, LDA, and PNN, achieved an 81.2% prediction rate for the 277 compounds.
• The consensus model outperformed individual classifiers in predicting the genotoxic class.

Conclusions:

• Computational models based on molecular structure descriptors can accurately predict PAC genotoxicity.
• The k-NN model and the developed consensus model offer efficient tools for genotoxicity assessment.
• These predictive approaches can aid in prioritizing chemicals for further testing and risk assessment.