Characteristic Substructures and Properties in the Chemical Carcinogenicity Studied by the Cascade Model

The cascade model is a rule induction methodology using the levelwise expansion of the lattice. An attribute-value pair is expressed as an item, and every node in the lattice is specified by an itemset and by its supporting instances. If the distribution of the class attribute values shows a large change along a link in the lattice, the link is represented as a rule "IF added-item-along-link added on itemset-on-upper-node, THEN class-i". The strength of the rule is measured by the BSS value of the link. In this study, we utilize linear substructure fragments and several physicochemical properties to describe a rule. A fragment leads to one of the two items [frag-i: y] and [frag-i: n] depending on whether or not the fragment exists in a molecule. Application of the cascade model to these items data set gives us rules about carcinogenicity. We could find several rules with large BSS values. Substructures and properties that appear in these rules are expected to provide a starting point for further chemical and biological study. Several rules with classification capability are used to predict the carcinogenicity for the compounds in the test set.