Statistical Modeling and Prediction of Carcinogenic Activity of Chemical Compounds

We have established two classification rules that statistically accurate allow to separate carcinogenically active chemical compounds from inactive chemical compounds. The electronic and information properties of molecules are used as molecular descriptors. The threshold values of descriptors that characterize and determine the presence or absence of carcinogenic properties of chemical compounds of various classes are found. Statistical quantitative indicators of the quality of classification rules are given, including the error of model. The proposed classification rules allow one to analyze the carcinogenic properties of different classes of chemical compounds from a unified view. Classification rules were tested for various classes of chemical compounds. We studied the chemical compounds of the following classes: a number of nitroso compounds, halogen-containing organic substances, sulfur-containing organic substances, aromatic amines and related compounds, dyes, oxy compounds, chemical compounds of the mustard type, and some medications. A total of 541 chemical compounds were examined.