QSAR for dihydrofolate reductase inhibitors with molecular graph structural descriptors

Abstract

Molecular graph descriptors are used, together with a large diversity of geometric, electrostatic, and quantum indices, to model physical, chemical, or biological properties with quantitative structure–property relationships and quantitative structure–activity relationships. The interest of developing new graph descriptors for organic compounds was stimulated in recent years by their use in virtual screening of combinatorial libraries, database mining, similarity and diversity assessment. Recently, we have extended topological indices by defining a series of molecular graph operators, providing an effective systematization and generalization of these structural descriptors. A graph operator uses a mathematical equation to compute a family of related molecular graph descriptors with different molecular matrices and various sets of parameters for atoms and bonds. In this paper we use structural descriptors computed with molecular graph operators to develop quantitative structure–activity relationships (QSAR) models for the dihydrofolate reductase inhibition with diaminopyrimidines. The molecular descriptors are derived from five molecular matrices, namely adjacency A, distance D, reciprocal distance RD, distance-path Dp, and reciprocal distance-path RDp. The QSAR models are obtained by selecting descriptors with a genetic algorithm, and the best models are validated with the leave-one-out cross-validation method. The QSAR models with the highest prediction power are comparable with those obtained with substituent constants and neural networks, but they use a much lower number of parameters.