CORAL: QSAR models for carcinogenicity of organic compounds for male and female rats

Abstract

Quantitative structure - activity relationships (QSARs) for carcinogenicity (rats, TD50) have been built up using the CORAL software. Different molecular features, which are extracted from simplified molecular input-line entry system (SMILES) serve as the basis for building up a model.Correlation weights for the molecular features are calculated by means of the Monte Carlo optimization. Using the numerical data on the correlation weights, one can calculate a model of carcinogenicity as a mathematical function of descriptors, which are sum of the corresponding correlation weights. In other words, the correlation weights provide the maximal correlation coefficient between the descriptor and carcinogenicity, for the training set. This correlation was assessed via external validation set. Finally, lists of molecular alerts in aspects of carcinogenicity for male rats and for female rats were compared and their differences were discussed.