Predicting Hydrophobic Solvation by Molecular Simulation – New United-Atom Model for Hydrocarbons

Solvation and partition properties assume fundamental importance in many fields, including pharmaceutical, oil and gas, environmental, and bioengineering [1]. Traditionally, such properties are predicted using empirical, group... [ view full abstract ]

Solvation and partition properties assume fundamental importance in many fields, including pharmaceutical, oil and gas, environmental, and bioengineering [1]. Traditionally, such properties are predicted using empirical, group contribution or statistical correlation approaches. More recently, however, molecular simulation-based methods have surfaced as a more rigorous alternative, and are assuming increasing importance due to their transferability and the continued increase in computational power [2]. In this context, development and validation of molecular models for solvation predictions becomes essential. We present a systematic test of available force-fields for solvation free energy predictions in non-aqueous solvents [3], starting with the simplest case – alkanes solvated in alkanes – and comparing three popular united-atom force-fields – OPLS-UA [4], GROMOS [5] and TraPPE [6]. Gibbs energies of solvation were calculated for 156 solute/solvent pairs from Molecular Dynamics (MD) simulations using Thermodynamic Integration. Our results show that, as the size of the alkanes increases, all models start to systematically deviate from experimental data [7]. An improved set of parameters is proposed for alkane molecules that corrects this systematic deviation and accurately predicts solvation free energies in hydrophobic media, while simultaneously providing a very good description of pure liquid densities and enthalpies of vaporization [8]. The model is then extended to alkenes and alkynes, again yielding very accurate predictions of solvation free energies and densities for these classes of compounds. The mean signed deviation from experimental data is very close to zero, indicating no systematic error in the predictions. The fact that predictions are robust even for relatively large molecules suggests that the new model may be applicable to solvation of non-polar macromolecules without accumulation of errors. The root mean squared deviation of the simulations is only 0.6 kJ/mol, which is lower than the estimated uncertainty in the experimental measurements. This excellent performance constitutes a solid basis upon which a more general model can be parameterized to describe solvation in both polar and non-polar environments.

[1]   Kollman, P. Chem. Rev. 93, 2395 (1993).

[2]   Chodera, J. D.; Mobley, D. L.; Shirts, M. R.; Dixon, R. W.; Branson, K.; Pande, V. S., Curr. Opin. Struct. Biol. 21, 150 (2011).

[3]   Jorge, M; Garrido, N. M.; Simões, C. J. V.; Silva, C. G.; Brito, R. M. M., J. Comput. Chem., 38, 346 (2017).

[4]   Jorgensen, W. L.; Madura, J. D.; Swenson, C. J., J. Am. Chem. Soc. 106, 6638 (1984).

[5]   Oostenbrink, C.; Villa, A.; Mark, A. E.; Van Gunsteren, W. F., J. Comp. Chem. 25, 1656 (2004).

[6]   Martin, M. G.; Siepmann, J. I., J. Phys. Chem. B 102, 2569 (1998).

[7]   Katritzky, A. R.; Tulp, I.; Fara, D. C.; Lauria, A.; Maran, U.; Acree, W. E. Jr, J. Chem. Inf. Model. 45, 913 (2005).

[8]   Jorge, M., J. Comput. Chem., 38, 359 (2017).