Data mining on carcinogenicity of chemical compounds by the JSM method

Prediction of the carcinogenicity of chemical compounds to rats was carried out by data mining analysis based on the logic of John Stuart Mill (JSM) and the fragmentary code of the substructure superposition (FCSS) data presentation language. The learning (608 compounds) and test (156 compounds) samples were taken from the database on the carcinogenicity of substances for laboratory animals developed by the Environmental Protection Agency of the USA (EPA USA). Predictions were made for 44% of the test samples. The prediction accuracy was 71%, sensitivity 73%, and specificity 67%. The causes for erroneous positive and negative predictions were studied, accounting for bioactivation of compounds in the course of metabolic biotransformations.