QSAR model based on weighted MCS trees approach for the representation of molecule data sets

In this paper we propose a new method for the generation of 2D-QSAR models for the prediction of activity values of chemicals. Maximum common substructures which are extracted from the data set are used for molecule classification in a tree, where the node of the tree represents molecules or common structures to groups of molecules and the arcs of the tree represent non isomorphic substructures between two nodes of the tree. All paths between pairwise leaf nodes are used to represent the equation system used as representational space in the building of the QSAR model. The proposed model, which is based on the combining of non isomorphic structures, use of molecular descriptors for the calculation of path lengths and classification of the data set based on maximum common substructures, considerably improves the generation of QSAR models with regard to the classical model based only on the use of a set of molecular descriptors. Optimization algorithms based on genetic algorithm and differential evolution approximations have also been used, resulting in the improvement and refinement of the equations obtained.