Application of a Fragment-based Model to the Prediction of the Genotoxicity of Aromatic Amines #
BioChem Press, Emilio Benfenati, Guido Sello · 2006
Motivation. Aromatic amines are well known mutagenic compounds, and their toxic effects have been thoroughly studied, thus making them a good test dataset for computational models. We developed a general approach to model compound toxicity, particularly to aquatic organisms. The model uses only diatomic fragments to estimate the compound biological response. We are now going to apply this model to a 95 compound dataset of aromatic amines to predict their genotoxicity. In particular, the computed activities are compared to the Salmonella mutagenicity tests that are a sufficiently homogeneous set of experimental data. Method. The method used is very straightforward. Each molecule is dissected into diatomic fragments; each of them is represented by atom type, interatomic bond type, and neighboring bond type. In this way, the variability of atomic fragments is well represented and guarantees a good representation of the molecular differences. Statistical analysis uses both standard multilinear regression and neural network analyses. Results. The initial dataset has been analyzed several times, using both the complete dataset and four partitions of this dataset in order to: (a) validate the model; (b) discuss the meaning of the result from a chemical viewpoint. The results are interesting because they demonstrate that our modeling approach is of general application and that the statistical analysis allows for the identification of some hints on the molecular characteristics that differentiate the biological activity of each compound. Conclusions. This new application of our modeling approach demonstrates that it could be used in several different models and applications.