Correction to Force Field Benchmark of Organic Liquids. 2. Gibbs Energy of Solvation

I a recent paper, we reported free energies of solvation ΔGsolv for organic solutes in organic solvents based on thermodynamic integration calculations, an empirical model due to Katritzky and co-workers and the conductor-like screening models for the realistic solvation (COSMO-RS) model. It was pointed out to us that the COSMO-RS results in our work seemed to be incorrect, which we confirmed and here we provide corrected values. By including an empirical correction for ring systems, the root-mean-square deviation from experiment for ΔGsolv for this method is reduced from 3.1 ± 0.2 to 2.3 ± 0.2 kJ/mol. Although the original paper stated the ring correction had been used, in fact it had not. A number of implementations of the semiempirical COSMO-RS method are available, all of which are slightly different. Here we used the version that can be bought as part of the Amsterdam Density Functional modeling suite. In our previous work, we neglected to use the empirical ring correction, which amounts to −0.887 kJ/mol per solute ring atom. This implies that all calculations with a ring system in the solute were too high, by 4.4 kJ/mol (5-membered rings) and 5.3 kJ/mol (6-membered rings), respectively. With this correction in place, the COSMO-RS method performs somewhat better than without, that is the root-mean-square deviation from experiment is reduced from 3.1 ± 0.2 to 2.3 ± 0.2 kJ/mol; however, the mean signed error changes from 0 to −1.2 ± 0.1 kJ/mol (see the Supporting Information and Figure 1). In addition, the correlation coefficient increases from 82% to 93%, from which we conclude that COSMO-RS yields better agreement with the experiment than either the QSPR method or the TI calculations. Indeed, a comparison of the correlation between different methods shows that the differences are significant (Table S1). Of the solvents used in the COSMO-RS method, dibutyl ether has the largest RMSD at 3.5 kJ/mol (N = 5) followed by chloroform (3.3 kJ/mol, N = 29). Although the BEDROC values and p-values of Student’s t-test of solvents suggest that chloroform (alkyl chloride, N = 29) and alcohol compounds are systematiclly off, the support is not strong because the BEDROC values are close to the BEDROCunif values that considers the uncertainty of BEDROC values, the p-values are close to the significance level 0.05, and the MSE is rather small. It should also be noted that a number of experimental data points are somewhat suspect because all three methods deviate in the same direction. For example, using chloroform as the solvent a deviation of ≈−10 kJ/mol from the experimental ΔGsolv is found for formaldehyde and (≈7 kJ/mol) for diethyl sulfide. Removing these data points only marginally affects the RMSD, and therefore they were left in. Solutes containing etherand carbonylgroups are systematically off as well, suggesting that the compounds with these functional groups are not described accurately by COSMO-RS.