The relationships between the capacity of chemicals to cumulate a toxic effect (functional cumulation) and the structure of their molecules were investigated. In the process of substantiation of safe levels (MAC) of substances in water this capacity is characterized by the cumulation hazard class (later in the text - hazard class). This class is stated to be depend on the value of the relationship between the mean lethal dose of the acute experiment and the threshold dose of the chronic experiment. The experimental study of a huge amount of new chemicals in the chronic experiments is a very difficult task, thus the study of the possibility to predict the hazard class of a chemical is of great scientific and practical interest. By using a logical combinatorial method JSM, named in honor of an English logic J.S. Mill, the structural groups in molecules, determining the appurtenance of these chemicals to a hazard class were identified and the possibility of the prediction of the hazard class of a chemical belonging to a definite structural array, containing such structural group were investigated. The training dataset (583 compounds) was automatically derived from the database WATERTOX, containing the data on acute and chronic toxicity for about 2000 substances. The results suggest the JSM method to be limitedly applicable for the determination of a hazard class of an untested chemical using this heterogeneous training dataset because we were unable to unambiguously derive the list of chemicals belonging to the class of moderately hazard substances. The chemical in some cases was predicted to belong to one or other of the neighboring classes. However taking in mind this uncertainty, the accuracy of the method evaluated, when using the “leave-one-out” method was 78%. Nevertheless the JSM method enables us to find structural subgroups “responsible” for the functional cumulation. The relation of the hazard class of a chemical belonging to a definite structural class with its structure and the possibility of the prediction of an untested chemical hazard class are demonstrated. The prognosis of the hazard classes for chemicals belonging to several structural sets including the anthraquinone derivatives, phthalimides, perfluorated aliphatic compounds, chlorosubstituted phenols, phenylureas is performed.
Program web-realization of JSM-method of automatic hypothesis generation based on inductive logic