Structure-mutagenicity modelling using counter propagation neural networks.

Area of Science:

• Computational chemistry
• Toxicology
• Machine learning

Background:

• Aromatic amines are a class of compounds with known mutagenic potential.
• Predicting mutagenicity is crucial for assessing chemical safety and guiding drug design.
• Quantitative Structure-Activity Relationship (QSAR) models are valuable tools for this prediction.

Purpose of the Study:

• To develop and validate a predictive model for the mutagenic potency of aromatic amines.
• To explore the utility of counter propagation neural networks (CPNN) for mutagenicity prediction.
• To identify structural similarities among aromatic amines using self-organising maps (SOMs).

Main Methods:

• A dataset of 95 aromatic amines and their mutagenic potencies was compiled.
• Compounds were characterized using four descriptor classes: topostructural (TS), topochemical (TC), geometrical, and quantum chemical (QC).
• A CPNN model was employed, incorporating SOMs for analysis and prediction, with leave-one-out (LOO) cross-validation for performance assessment.

Main Results:

• The developed model demonstrated predictive capability with squared correlation coefficients ranging from 0.65 to 0.75.
• The model's performance was comparable to traditional linear modeling approaches.
• Analysis of the self-organising maps revealed distinct clusters of structurally related compounds.

Conclusions:

• Counter propagation neural networks are effective for predicting the mutagenicity of aromatic amines.
• The study successfully identified structure-mutagenicity relationships and grouped similar compounds.
• This approach aids in understanding chemical toxicity and designing safer molecules.