Predictive QSPR study of the dissociation constants of diverse pharmaceutical compounds

Abstract

The objective of the article was to perform a predictive analysis, based on quantitative structure-property relationships, of the dissociation constants (pKa) of different medicinal compounds (e.g., salicylic acid, salbutamol, lidocaine). Given the importance of this property in medicinal chemistry, it is of interest to develop theoretical methods for its prediction. The descriptors selection from a pool containing more than a thousand geometrical, topological, quantum-mechanical, and electronic types of descriptors was performed using the enhanced replacement method. Genetic algorithm and the replacement method (RM) techniques were used as reference points. A new methodology for the selection of the optimal number of descriptors to include in a model was presented and successfully used, showing that the best model should contain four descriptors. The best quantitative structure-property relationships linear model constructed using 62 molecular structures not previously used in this type of quantitative structure-property study showed good predictive attributes. The root mean squared error of the 26 molecules test set was 0.5600. The analysis of the quantitative structure-property relationships model suggests that the dissociation constants depend significantly on the number of acceptor atoms for H-bonds and on the number of carboxylic acids present in the molecules.