QSPR correlation of melting point for drug compounds based on different sources of molecular descriptors.

Five linear QSPR models for melting points (MP) of drug-like compounds are developed based on three different packages for molecular descriptor generation and a combined set of all descriptors. A data set of 323 gaseous, liquid, and solid compounds was used for this study. Two models from the combined set of descriptors based on stepwise regression and genetic algorithm (GA) descriptor selection methods have acceptable prediction abilities. The statistical results of these models are r2 = 0.673 and root-mean-square error (RMSE) of 40.4 degrees C for stepwise regression-based quantitative structure-property relationships (QSPRs) and r2 = 0.660 and RMSE of 41.1 degrees C for GA-based QSPRs. Interpretation of descriptors of all models showed a strong correlation of hydrogen bonding and molecular complexity with melting points of drug-like compounds.