QSPR modeling of lipid-water partition coefficient by optimization of correlation weights of local graph invariants.

The optimization of correlation weights scheme was applied to model lipid-water partition coefficient (log P) of two sets of diverse functional aliphatic and aromatic compounds. In both cases, the optimized descriptors formulated based on the data of training sets generated statistically acceptable relations for the corresponding training sets, test sets, and combined sets. When the relations of log P values with the optimized molecular descriptors formulated based on the data of the training sets were used for calculation of log P values of the corresponding training sets, rpred2 values were found to be satisfactory (above 0.99) in both cases, which is indicative of the predictive potential of the scheme. The results indicate promising potential of the optimization of a correlation weights scheme in modeling studies.