Free energies of solvation in the context of protein folding: Implications for implicit and explicit solvent models

Conformational energies of biomolecules in solution: Extending the MPCONF196 benchmark with explicit water molecules

A prerequisite for the computational prediction of molecular properties like conformational energies of biomolecules is a reliable, robust, and computationally affordable method usually selected according to its performance for relevant benchmark sets. However, most of these sets comprise molecules in the gas phase and do not cover interactions with a solvent, even though biomolecules typically occur in aqueous solution. To address this issue, we introduce a with explicit water molecules solvated version of a gas-phase benchmark set containing 196 conformers of 13 peptides and other relevant macrocycles, namely MPCONF196 [J. Řezáč et al., JCTC 2018, 14, 1254-1266], and provide very accurate PNO-LCCSD(T)-F12b/AVQZ' reference values. The novel solvMPCONF196 benchmark set features two additional challenges beyond the description of conformers in the gas phase: conformer-water and water-water interactions. The overall best performing method for this set is the double hybrid revDSDPBEP86-D4/def2-QZVPP yielding conformational energies of almost coupled cluster quality. Furthermore, some (meta-)GGAs and hybrid functionals like B97M-V and ω B97M-D with a large basis set reproduce the coupled cluster reference with an MAD below 1 kcal mol- 1  . If more efficient methods are required, the composite DFT-method r2 SCAN-3c (MAD of 1.2 kcal mol- 1 ) is a good alternative, and when conformational energies of polypeptides or macrocycles with more than 500-1000 atoms are in the focus, the semi-empirical GFN2-xTB or the MMFF94 force field (for very large systems) are recommended.

Enhanced sampling in explicit solvent by deep learning module in FSATOOL

FSATOOL is an integrated molecular simulation and data analysis program. Its old molecular dynamics engine only supports simulations in vacuum or implicit solvent. In this work, we implement the well-known smooth particle mesh Ewald method for simulations in explicit solvent. The new developed engine is runnable on both CPU and GPU. All the existed analysis modules in the program are compatible with the new engine. Moreover, we also build a complete deep learning module in FSATOOL. Based on the module, we further implement two useful trajectory analysis methods: state-free reversible VAMPnets and time-lagged autoencoder. They are good at searching the collective variables related to the conformational transitions of biomolecules. In FSATOOL, these collective variables can be further used to construct a bias potential for the enhanced sampling purpose. We introduce the implementation details of the methods and present their actual performances in FSATOOL by a few enhanced sampling simulations.

Machine learning based implicit solvent model for aqueous-solution alanine dipeptide molecular dynamics simulations

Inspired by the recent work from Noé and coworkers on the development of machine learning based implicit solvent model for the simulation of solvated peptides [Chen et al., J. Chem. Phys., 2021, 155, 084101], here we report another investigation of the possibility of using machine learning (ML) techniques to "derive" an implicit solvent model directly from explicit solvent molecular dynamics (MD) simulations. For alanine dipeptide, a machine learning potential (MLP) based on the DeepPot-SE representation of the molecule was trained to capture its interactions with its average solvent environment configuration (ASEC). The predicted forces on the solute deviated only by an RMSD of 0.4 kcal mol-1 Å-1 from the reference values, and the MLP-based free energy surface differed from that obtained from explicit solvent MD simulations by an RMSD of less than 0.9 kcal mol-1. Our MLP training protocol could also accurately reproduce combined quantum mechanical molecular mechanical (QM/MM) forces on the quantum mechanical (QM) solute in ASEC environment, thus enabling the development of accurate ML-based implicit solvent models for ab initio-QM MD simulations. Such ML-based implicit solvent models for QM calculations are cost-effective in both the training stage, where the use of ASEC reduces the number of data points to be labelled, and the inference stage, where the MLP can be evaluated at a relatively small additional cost on top of the QM calculation of the solute.

Symmetrization in the Calculation Pipeline of Gauss Function-Based Modeling of Hydrophobicity in Protein Structures

In this paper, we show, discuss, and compare the effects of symmetrization in two calculation subroutines of the Fuzzy Oil Drop model, a coarse-grained model of density of hydrophobicity in proteins. In the FOD model, an input structure is enclosed in an axis-aligned ellipsoid called a drop. Two profiles of hydrophobicity are then calculated for its residues: theoretical (based on the 3D Gauss function) and observed (based on pairwise hydrophobic interactions). Condition of the hydrophobic core is revealed by comparing those profiles through relative entropy, while analysis of their local differences allows, in particular, determination of the starting location for the search for protein–protein and protein–ligand interaction areas. Here, we improve the baseline workflow of the FOD model by introducing symmetry to the hydrophobicity profile comparison and ellipsoid bounding procedures. In the first modification (FOD–JS), Kullback–Leibler divergence is enhanced with its Jensen–Shannon variant. In the second modification (FOD-PCA), the molecule is optimally aligned with the axes of the coordinate system via principal component analysis, and the size of its drop is determined by the standard deviation of all its effective atoms, making it less susceptible to structural outliers. Tests on several molecules with various shapes and functions confirm that the proposed modifications improve the accuracy, robustness, speed, and usability of Gauss function-based modeling of the density of hydrophobicity in protein structures.

Machine learning implicit solvation for molecular dynamics

Accurate modeling of the solvent environment for biological molecules is crucial for computational biology and drug design. A popular approach to achieve long simulation time scales for large system sizes is to incorporate the effect of the solvent in a mean-field fashion with implicit solvent models. However, a challenge with existing implicit solvent models is that they often lack accuracy or certain physical properties compared to explicit solvent models as the many-body effects of the neglected solvent molecules are difficult to model as a mean field. Here, we leverage machine learning (ML) and multi-scale coarse graining (CG) in order to learn implicit solvent models that can approximate the energetic and thermodynamic properties of a given explicit solvent model with arbitrary accuracy, given enough training data. Following the previous ML-CG models CGnet and CGSchnet, we introduce ISSNet, a graph neural network, to model the implicit solvent potential of mean force. ISSNet can learn from explicit solvent simulation data and be readily applied to molecular dynamics simulations. We compare the solute conformational distributions under different solvation treatments for two peptide systems. The results indicate that ISSNet models can outperform widely used generalized Born and surface area models in reproducing the thermodynamics of small protein systems with respect to explicit solvent. The success of this novel method demonstrates the potential benefit of applying machine learning methods in accurate modeling of solvent effects for in silico research and biomedical applications.

An ABSINTH-Based Protocol for Predicting Binding Affinities between Proteins and Small Molecules

The core task in computational drug discovery is to accurately predict binding free energies in receptor-ligand systems for large libraries of putative binders. Here, the ABSINTH implicit solvent model and force field are extended to describe small, organic molecules and their interactions with proteins. We show that an automatic pipeline based on partitioning arbitrary molecules into substructures corresponding to model compounds with known free energies of solvation can be combined with the CHARMM general force field into a method that is successful at the two important challenges a scoring function faces in virtual screening work flows: it ranks known binders with correlation values rivaling that of comparable state-of-the-art methods and it enriches true binders in a set of decoys. Our protocol introduces innovative modifications to common virtual screening workflows, notably the use of explicit ions as competitors and the integration over multiple protein and ligand species differing in their protonation states. We demonstrate the value of modifications to both the protocol and ABSINTH itself. We conclude by discussing the limitations of high-throughput implicit methods such as the one proposed here.

Sequence Tolerance of a Single-Domain Antibody with a High Thermal Stability: Comparison of Computational and Experimental Fitness Profiles

The sequence fitness of a llama single-domain antibody with an unusually high thermal stability is explored by a combined computational and experimental study. Starting with the X-ray crystallographic structure, RosettaBackrub simulations were applied to model sequence-structure tolerance profiles and identify key substitution sites. From the model calculations, an experimental site-directed mutagenesis was used to produce a panel of mutants, and their melting temperatures were determined by thermal denaturation. The results reveal a sequence fitness of an excess stability of approximately 12 °C, a value taken from a decrease in the melting temperature of an electrostatic charge-reversal substitution in the CRD3 without a deleterious effect on the binding affinity to the antigen. The tolerance for the disruption of antigen recognition without loss in the thermal stability was demonstrated by the introduction of a proline in place of a tyrosine in the CDR2, producing a mutant that eliminated binding. To further assist the sequence design and the selection of engineered single-domain antibodies, an assessment of different computational strategies is provided of their accuracy in the detection of substitution "hot spots" in the sequence tolerance landscape.

Computational Estimation of Microsecond to Second Atomistic Folding Times

Despite the development of massively parallel computing hardware including inexpensive graphics processing units (GPUs), it has remained infeasible to simulate the folding of atomistic proteins at room temperature using conventional molecular dynamics (MD) beyond the microsecond scale. Here, we report the folding of atomistic, implicitly solvated protein systems with folding times τ ranging from ∼10 μs to ∼100 ms using the weighted ensemble (WE) strategy in combination with GPU computing. Starting from an initial structure or set of structures, WE organizes an ensemble of GPU-accelerated MD trajectory segments via intermittent pruning and replication events to generate statistically unbiased estimates of rate constants for rare events such as folding; no biasing forces are used. Although the variance among atomistic WE folding runs is significant, multiple independent runs are used to reduce and quantify statistical uncertainty. Folding times are estimated directly from WE probability flux and from history-augmented Markov analysis of the WE data. Three systems were examined: NTL9 at low solvent viscosity (yielding τf = 0.8-9 μs), NTL9 at water-like viscosity (τf = 0.2-2 ms), and Protein G at low viscosity (τf = 3-200 ms). In all cases, the folding time, uncertainty, and ensemble properties could be estimated from WE simulation; for Protein G, this characterization required significantly less overall computing than would be required to observe a single folding event with conventional MD simulations. Our results suggest that the use and calibration of force fields and solvent models for precise estimation of kinetic quantities is becoming feasible.

Improvements to the ABSINTH Force Field for Proteins Based on Experimentally Derived Amino Acid Specific Backbone Conformational Statistics

We present an improved version of the ABSINTH implicit solvation model and force field paradigm (termed ABSINTH-C) by incorporating a grid-based term that bootstraps against experimentally derived and computationally optimized conformational statistics for blocked amino acids. These statistics provide high-resolution descriptions of the intrinsic backbone dihedral angle preferences for all 20 amino acids. The original ABSINTH model generates Ramachandran plots that are too shallow in terms of the basin structures and too permissive in terms of dihedral angle preferences. We bootstrap against the reference optimized landscapes and incorporate CMAP-like residue-specific terms that help us reproduce the intrinsic dihedral angle preferences of individual amino acids. These corrections that lead to ABSINTH-C are achieved by balancing the incorporation of the new residue-specific terms with the accuracies inherent to the original ABSINTH model. We demonstrate the robustness of ABSINTH-C through a series of examples to highlight the preservation of accuracies as well as examples that demonstrate the improvements. Our efforts show how the recent experimentally derived and computationally optimized coil-library landscapes can be used as a touchstone for quantifying errors and making improvements to molecular mechanics force fields.

Assessing AMBER force fields for protein folding in an implicit solvent

Molecular dynamics (MD) simulation implemented with a state-of-the-art protein force field and implicit solvent model is an attractive approach to investigate protein folding, one of the most perplexing problems in molecular biology. But how well can force fields developed independently of implicit solvent models work together in reproducing diverse protein native structures and measuring the corresponding folding thermodynamics is not always clear. In this work, we performed enhanced sampling MD simulations to assess the ability of six AMBER force fields (FF99SBildn, FF99SBnmr, FF12SB, FF14ipq, FF14SB, and FF14SBonlysc) as coupled with a recently improved pair-wise GB-Neck2 model in modeling the folding of two helical and two β-sheet peptides. Whilst most of the tested force fields can yield roughly similar features for equilibrium conformational ensembles and detailed folding free-energy profiles for short α-helical TC10b in an implicit solvent, the measured counterparts are significantly discrepant in the cases of larger or β-structured peptides (HP35, 1E0Q, and GTT). Additionally, the calculated folding/unfolding thermodynamic quantities can only partially match the experimental data. Although a combination of the force fields and GB-Neck2 implicit model able to describe all aspects of the folding transitions towards the native structures of all the considered peptides was not identified, we found that FF14SBonlysc coupled with the GB-Neck2 model seems to be a reasonably balanced combination to predict peptide folding preferences.

The effects of implicit modeling of nonpolar solvation on protein folding simulations

Implicit solvent models, in which the polar and nonpolar solvation free-energies of solute molecules are treated separately, have been widely adopted for molecular dynamics simulation of protein folding. While the development of the implicit models is mainly focused on the methodological improvement and key parameter optimization for polar solvation, nonpolar solvation has been either ignored or described by a simplistic surface area (SA) model. In this work, we performed the folding simulations of multiple β-hairpin and α-helical proteins with varied surface tension coefficients embedded in the SA model to clearly demonstrate the effects of nonpolar solvation treated by a popular SA model on protein folding. The results indicate that the change in the surface tension coefficient does not alter the ability of implicit solvent simulations to reproduce a protein native structure but indeed controls the components of the equilibrium conformational ensemble and modifies the energetic characterization of the folding transition pathway. The suitably set surface tension coefficient can yield explicit solvent simulations and/or experimentally suggested folding mechanism of protein. In addition, the implicit treatment of both polar and nonpolar components of solvation free-energy contributes to the overestimation of the secondary structure in implicit solvent simulations.

Explicit Solvent Hydration Benchmark for Proteins with Application to the PBSA Method

Explicit and implicit solvent models have a proven record of delivering hydration free energies of small, druglike solutes in reasonable agreement with experiment. Hydration of macromolecules, such as proteins, is to a large extent uncharted territory, with few results shedding light on quantitative consistency between different solvent models, let alone their ability to reproduce real water. In this work, based on extensive explicit solvent simulations employing TIP3P and SPC/E water models we analyze hydration free energy changes between fixed conformations of 5 diverse proteins, including large multidomain structures. For the two solvent models we find better agreement in electrostatic rather than nonpolar contributions (RMSE of 2.3 and 2.7 kcal/mol, respectively), even though absolute values of the latter are typically an order of magnitude smaller. We also highlight the importance of finite size corrections to relative protein hydration free energies, which turn out to be rather large, on the order of several kcal/mol, and are necessary for proper interpretation of results obtained under periodic boundary conditions. We further compare gathered data with predictions of the implicit solvent approach based on the Poisson equation and the surface or volume based nonpolar term. We find definitely lesser consistency than between the two explicit models (RMSE between implicit and TIP3 results of 11.3 and 8.4 kcal/mol for electrostatic and nonpolar contributions, respectively). In the process we determine the value of the protein dielectric constant and the geometric model for the dielectric boundary that provide for the best agreement. Finally, we evaluate the usefulness of surface and volume based models of nonpolar contributions to hydration free energy of large biomolecules.