Accurate prediction of hydration free energies and solvation structures using molecular density functional theory with a simple bridge functional

Daniel Borgis, Sohvi Luukkonen, Luc Belloni, Guillaume JeanmairetThe Journal of Chemical Physics 2021 · DOI

TL;DR

This paper introduces a refined molecular density functional theory (MDFT) that accurately predicts hydration free energies and solvent structures for a wide range of molecules. By incorporating a novel "bridge functional" into existing approximations, the method achieves near-experimental accuracy for neutral molecules with a significant speedup over traditional simulations, paving the way for more efficient molecular modeling.

What Problem Does This Paper Address?

Predicting how molecules interact with water – specifically, their "hydration free energy" (HFE) and how water molecules arrange themselves around them – is crucial for understanding many chemical and biological processes. Traditional methods like molecular dynamics (MD) or Monte Carlo (MC) simulations can be highly accurate but are computationally very expensive, often requiring weeks or months of computation for a single HFE calculation. Existing theoretical approaches, like implicit solvent models, are faster but lack the detail to capture the microscopic solvent structure. Molecular density functional theory (MDFT) offers a promising middle ground, but previous iterations (like the HNC approximation) struggled with accuracy, particularly for hydrophobic and hydrophilic interactions, and often required empirical corrections. This paper aims to bridge this gap by developing a more accurate and efficient MDFT that can reliably predict both HFEs and solvent structures for a broad spectrum of molecules.

Methods

The core of this work lies in enhancing the Molecular Density Functional Theory (MDFT) framework. MDFT treats the solvent (in this case, water) as a continuous field whose properties are described by its density at each point in space and orientation. The theory aims to minimize a free energy functional to find the equilibrium solvent configuration.

The authors build upon the hyper-netted chain (HNC) approximation, which is a second-order approximation of the solvent's excess free energy. They introduce a "bridge functional" (FB), which accounts for higher-order contributions (beyond the second order) that are neglected in HNC. Specifically, they propose a simple, angular-independent Weighted Density Approximation (WDA) bridge functional. This functional is designed to be parameter-free, relying only on fundamental bulk water properties (compressibility, surface tension, etc.) that are independently determined.

To further improve accuracy for hydrophilic solutes, a simple angular-dependent correction is added to account for electrostatic interactions. This modified functional is termed HNC+WDA*.

For comparison and validation, the authors employ hybrid-4th-dimension Monte-Carlo (H4D-MC) simulations, a computationally efficient simulation method, to generate reference data for various solutes.

Datasets

The paper utilizes several datasets for evaluation:

  • Hard-sphere solutes: Used to test the basic hydrophobic solvation model.
  • Lennard-Jones solutes: Used to assess the inclusion of attractive forces.
  • Hydrophobic pairs (coronene molecules): To study hydrophobic interactions.
  • Water as a solute: To specifically test the model's ability to describe water-water interactions and calibrate electrostatic corrections.
  • Reduced dataset of 19 organic molecules: A curated set including amino-acid side-chain analogs and other molecules to test hydrophilic solvation.
  • FreeSolv database: A large dataset of 642 drug-like molecules for which experimental and simulation-based hydration free energies are available. This dataset is used to evaluate the overall predictive power of the HNC+WDA* functional for neutral molecules.

Key Insights and Ideas

  • The power of the bridge functional: The introduction of a simple, parameter-free bridge functional significantly improves the accuracy of MDFT, particularly for hydrophobic solvation, by capturing higher-order correlations in the solvent.
  • Bridging the gap between theory and simulation: The HNC+WDA functional achieves a remarkable balance between the accuracy of expensive simulations and the speed of theoretical methods, making it a viable tool for larger-scale studies.
  • Parameter-free approach: The fact that the functional is parameterized solely on bulk water properties, without needing to be tuned to specific solute data, is a major strength, indicating a more fundamental and transferable model.
  • Limitations for charged species: While successful for neutral molecules, the current MDFT framework still struggles to accurately predict the solvation of ions, highlighting the complexity of electrostatic interactions in water.
  • Water's unique complexity: The paper underscores that accurately modeling liquid water remains a significant challenge, even with advanced theoretical tools.

Key Results

  • Hydrophobic Solvation: The HNC+WDA functional accurately predicts hydration free energies and solvent structures for hard-sphere and Lennard-Jones solutes, outperforming previous MDFT approximations and even some simulation-based corrections.
  • FreeSolv Database: For the large FreeSolv dataset of neutral molecules, the HNC+WDA functional achieves a mean absolute error (MAE) of 0.45 kBT* (approximately 1 kJ/mol), which is considered chemical accuracy and a significant improvement over previous methods.
  • Speedup: The MDFT approach offers a speedup of 3 orders of magnitude compared to traditional simulation methods for calculating hydration free energies.
  • Solvent Structure: The method accurately reproduces key features of the solvent structure around solutes, such as radial distribution functions, which are crucial for understanding solvation.
  • Charged Species: For monovalent ions (K+ and Cl-), the HNC+WDA* functional shows some improvement over HNC but still underestimates solvation free energies, indicating room for further development.

Limitations and Future Work

The authors acknowledge several limitations:

  • Charged Species: The current MDFT framework, even with the electrostatic correction, does not accurately predict the hydration free energies of ionic species. The parameter used for the electrostatic correction needs to be adjusted for different charges and types of ions.
  • H-bonding Structure: While improved, the description of the H-bonding structure, particularly around anions, still requires refinement.
  • Future Directions: The authors suggest that a more systematic, angular-dependent bridge functional, potentially incorporating three-body correlations, could lead to further improvements, especially for charged species and a more accurate representation of water's H-bonding network. They also envision applying this refined MDFT to complex chemical problems, including biomolecular interactions and QM/MM approaches.

Summary updated

Abstract

This paper assesses the ability of molecular density functional theory to predict efficiently and accurately the hydration free energies of molecular solutes and the surrounding microscopic water structure. A wide range of solutes were investigated, including hydrophobes, water as a solute, and the FreeSolv database containing 642 drug-like molecules having a variety of shapes and sizes. The usual second-order approximation of the theory is corrected by a third-order, angular-independent bridge functional. The overall functional is parameter-free in the sense that the only inputs are bulk water properties, independent of the solutes considered. These inputs are the direct correlation function, compressibility, liquid–gas surface tension, and excess chemical potential of the solvent. Compared to molecular simulations with the same force field and the same fixed solute geometries, the present theory is shown to describe accurately the solvation free energy and structure of both hydrophobic and hydrophilic solutes. Overall, the method yields a precision of order 0.5 kBT for the hydration free energies of the FreeSolv database, with a computer speedup of 3 orders of magnitude. The theory remains to be improved for a better description of the H-bonding structure and the hydration free energy of charged solutes.
BibTeX
@article{borgis2021accurate,
  author = {Daniel Borgis and Sohvi Luukkonen and Luc Belloni and Guillaume Jeanmairet},
  title = {{Accurate prediction of hydration free energies and solvation structures using molecular density functional theory with a simple bridge functional}},
  journal = {The Journal of Chemical Physics},
  year = {2021},
  doi = {10.1063/5.0057506},
}
Figure