In the discovery process of new drugs and the development of novel therapies in medicine, computational modeling is a complementary tool for the design of new molecules by predicting for example their solubility in different solvents. Here, we benchmarked several computational methods to calculate the partition coefficients of a diverse set of 161 organic molecules with experimental logP values obtained from the literature. In general, density functional theory methods yielded the best correlations and lower average deviations. Although results are obtained faster with semiempirical and molecular mechanics methodologies, these methods yielded higher average deviations and lower correlation coefficients than hybrid density functional theory methods. We recommend the use of an empirical formula to correct the calculated values with each methodology tested.

In the discovery process of new drugs and the development of novel therapies in medicine, computational modeling is a complementary tool for the design of new molecules by predicting for example their solubility in different solvents. Here, we benchmarked several computational methods to calculate the partition coefficients of a diverse set of 161 organic molecules with experimental logP values obtained from the literature. In general, density functional theory methods yielded the best correlations and lower average deviations. Although results are obtained faster with semiempirical and molecular mechanics methodologies, these methods yielded higher average deviations and lower correlation coefficients than hybrid density functional theory methods. We recommend the use of an empirical formula to correct the calculated values with each methodology tested.