How accurate are the popular PCM/GB continuum solvation models for calculating the solvation energies of amino acid side-chain analogs?

Implicit/continuum solvation models are effective methods that are widely used to account for sol-vation effects. Because a large number of empirical parameters are used in these models, it is of significance to identify proper parameter sets. To assess the performance of the popular polarizable continuum models in Gaussian 03 (G03) and 09 (G09) and generalized born (GB) models in AMBER 11, we have computed the solvation energies of fifteen neutral amino acid side-chain analogs at various levels by systematically varying parameters (over 2,668 sets of calculations). The evaluation using the experimental values as standards leads to the following observations: (1) among all the tested methods, IEFPCM/UAKS rather than the default IEFPCM/UA0 in G03 performs best with a 0.21 ± 0.21 kcal/mol of mean absolute deviation ± standard deviation of unsigned errors (MAD ± SD). Unexpectedly, the default IEFPCM newly implemented in G09 performs poorly. Detailed analyses reveal that the electrostatic contribution was not accounted properly, due to changing the default Alpha scaling factor from 1.2 in G03 to 1.1 in G09. When setting the factor back to 1.2, the G09 IEFPCM with a continuous surface charge model performs comparably to the G03 IEFPCM with a point surface charge model. (2) The SMD model performs well in G09 but slightly less accurate than the IEFPCM/UAKS in G03 by ~ 0.1 kcal/mol of MAD. (3) In AMBER 11, when the atomic partial charges derived from the commonly used HF/6-31G* electrostatic potentials are used, GB7 in combination with mBondi2 radii with 1.01 ± 0.67 kcal/mol of MAD ± SD performs better than the combinations of other GB methods and radii. However, GB8/Bondi, when using the charges derived from MP2/6-311++G** calculations, performs best with 0.78 ± 0.58 kcal/mol of MAD ± SD) among all GB calculations. (4) The use of the charges, derived from QM calculations in the condense phase, does not improve the performance, indicating that reoptimiza-tion of GB parameters is required for using the condense phase charges.