Quantitative structure-activity relationship models predict mutagenicity in 95 aromatic amines, indicating the effectiveness of new graph-based descriptors.
Key Points
The study aims to develop predictive QSAR models to assess the mutagenicity of aromatic and heteroaromatic amines.
Developed QSAR models using Laplacian matrix-based molecular descriptors.
Utilized graph convolution to calculate molecular properties from atom and structure information.
Employed counter-propagation artificial neural network optimized with a genetic algorithm.
The new descriptors effectively encoded structural information for predicting TA98 Ames’ mutagenicity.
Demonstrated significant predictive capability for the mutagenicity of the tested amines.