Advancements and Applications of Molecular Dynamics Simulations in Biological Research: A Comprehensive Review

Many thanks to our sponsor Esdebe who helped us prepare this research report.

Abstract

Molecular dynamics (MD) simulations have emerged as a pivotal tool in modern biological research, providing atomistic-level insights into the structure, dynamics, and interactions of biomolecules. This review offers a comprehensive overview of the principles, methodologies, and applications of MD simulations, with a particular focus on recent advancements and their impact on understanding complex biological processes. We delve into the underlying theoretical framework, force fields, simulation algorithms, and enhanced sampling techniques that are essential for accurate and efficient MD simulations. Furthermore, we discuss the role of MD simulations in various biological domains, including protein folding, enzyme catalysis, drug discovery, membrane dynamics, and nucleic acid structure-function relationships. We also address the limitations of MD simulations, such as computational cost and force field accuracy, and highlight strategies for integrating MD simulations with experimental data to obtain a more holistic understanding of biological systems. Finally, we discuss the future directions of MD simulations and their potential to further transform biological research.

Many thanks to our sponsor Esdebe who helped us prepare this research report.

1. Introduction

The past few decades have witnessed a remarkable revolution in our ability to study biological systems at the molecular level. This revolution is fueled, in part, by the synergistic development of experimental and computational methodologies. Among the computational techniques, Molecular Dynamics (MD) simulation stands out as a powerful tool to investigate the atomic-level behavior of macromolecules, such as proteins, nucleic acids, lipids, and carbohydrates [1]. MD simulations use classical mechanics to model the time evolution of a system by solving Newton’s equations of motion for each atom, allowing researchers to observe the dynamic behavior of biomolecules over time. This ability to simulate biological processes at the atomic level provides insights that are often difficult or impossible to obtain from experiments alone.

MD simulations have become indispensable in numerous areas of biological research. In protein science, MD is used to study protein folding pathways, protein-protein interactions, enzyme mechanisms, and the effects of mutations on protein structure and function [2, 3]. In drug discovery, MD simulations are used to screen potential drug candidates, optimize drug binding affinity, and understand drug resistance mechanisms [4]. MD is also crucial in understanding the structure and dynamics of biological membranes [5], and in unraveling the complex behavior of nucleic acids [6].

Despite its power, MD simulation also faces several challenges. The accuracy of MD simulations relies heavily on the quality of the force field used to describe the interactions between atoms. Furthermore, simulating biological processes that occur over long timescales, such as protein folding or large conformational changes, can be computationally demanding. Overcoming these challenges requires the development of more accurate force fields, efficient simulation algorithms, and advanced sampling techniques. In this review, we will explore these challenges and the recent advancements made to address them, highlighting the increasing role of MD simulations in driving biological discoveries.

Many thanks to our sponsor Esdebe who helped us prepare this research report.

2. Theoretical Foundations and Methodologies

2.1 The Fundamentals of Molecular Dynamics

At the heart of MD simulations lies the application of classical mechanics to describe the motion of atoms. Given an initial set of atomic coordinates and velocities, MD simulations iteratively solve Newton’s equations of motion, F = ma, where F is the force acting on an atom, m is its mass, and a is its acceleration. The force on each atom is calculated from a potential energy function, often referred to as a force field, which describes the interactions between atoms. By integrating Newton’s equations over time using numerical algorithms, MD simulations generate a trajectory of the atomic positions and velocities, providing a dynamic view of the system’s evolution [7].

2.2 Force Fields

The accuracy of an MD simulation hinges on the quality of the force field. A force field is a set of mathematical equations and parameters that define the potential energy of the system as a function of atomic coordinates. Most commonly used force fields in biomolecular simulations are additive, meaning they describe the total potential energy as a sum of bonded and non-bonded interactions. Bonded interactions include terms for bond stretching, angle bending, and dihedral angle torsion. Non-bonded interactions consist of van der Waals interactions (typically modeled using the Lennard-Jones potential) and electrostatic interactions (modeled using Coulomb’s law) [8].

Several widely used force fields have been developed for biomolecular simulations, including AMBER, CHARMM, GROMOS, and OPLS [9-12]. Each force field has its own strengths and weaknesses, and the choice of force field often depends on the specific application. For instance, AMBER and CHARMM are frequently used for protein and nucleic acid simulations, while GROMOS is sometimes favored for simulations of systems with high solvent content. Recently, polarizable force fields have emerged as a more sophisticated approach to account for the electronic polarization effects that are not captured by traditional additive force fields [13]. These polarizable force fields, such as AMOEBA and Drude, explicitly model the induced dipoles on atoms, leading to a more accurate representation of electrostatic interactions. However, polarizable force fields are computationally more expensive than additive force fields.

2.3 Simulation Algorithms and Ensembles

MD simulations require numerical algorithms to integrate Newton’s equations of motion over time. The Verlet algorithm and its variants (e.g., leap-frog Verlet, velocity Verlet) are commonly used due to their simplicity, stability, and time-reversibility [14]. The time step used in MD simulations must be small enough to accurately capture the fastest motions in the system, typically on the order of femtoseconds (10^-15 s). As a result, simulating biological processes that occur on longer timescales requires significant computational resources.

MD simulations are typically performed in specific thermodynamic ensembles, which define the conditions under which the system is simulated. The most common ensembles are the microcanonical ensemble (NVE, constant number of particles, volume, and energy), the canonical ensemble (NVT, constant number of particles, volume, and temperature), and the isothermal-isobaric ensemble (NPT, constant number of particles, pressure, and temperature). The NVT and NPT ensembles require the use of thermostats and barostats, respectively, to maintain the temperature and pressure of the system. Several thermostatting and barostatting algorithms are available, such as the Berendsen thermostat/barostat, the Nosé-Hoover thermostat/barostat, and the Langevin thermostat [15-17].

2.4 Enhanced Sampling Techniques

A significant limitation of MD simulations is the difficulty in sampling rare events, such as protein folding or conformational transitions, which occur on timescales much longer than the typical simulation time. Enhanced sampling techniques have been developed to overcome this limitation by accelerating the exploration of the system’s energy landscape. These techniques can be broadly classified into two categories: biased and unbiased methods [18].

Biased methods introduce a bias potential to the system to encourage sampling of specific regions of the energy landscape. Examples of biased methods include umbrella sampling, metadynamics, and accelerated molecular dynamics (AMD) [19-21]. Umbrella sampling divides the reaction coordinate into windows and applies a harmonic potential to constrain the system within each window. Metadynamics adds a history-dependent potential to discourage the system from revisiting previously sampled states. AMD reduces the energy barriers by modifying the potential energy surface.

Unbiased methods, such as replica exchange molecular dynamics (REMD), also known as parallel tempering, enhance sampling by running multiple simulations at different temperatures and periodically exchanging configurations between replicas [22]. REMD allows the system to overcome energy barriers more easily at higher temperatures and then transfer the sampled configurations to lower temperatures. Another unbiased method is Markov state modeling (MSM), which combines short MD simulations with statistical analysis to construct a kinetic model of the system’s dynamics [23].

Many thanks to our sponsor Esdebe who helped us prepare this research report.

3. Software Packages and Computational Resources

Several software packages are available for performing MD simulations, each with its own strengths and features. Some of the most widely used packages include:

• GROMACS: A versatile and highly efficient package particularly well-suited for simulations of proteins, lipids, and nucleic acids [24]. GROMACS is known for its excellent parallelization capabilities, making it ideal for large-scale simulations on high-performance computing clusters.
• NAMD: A parallel MD code designed for high-performance simulation of large biomolecular systems [25]. NAMD is particularly strong in its support for steered molecular dynamics (SMD) and free energy perturbation (FEP) calculations.
• AMBER: A suite of programs for biomolecular simulations, including the Sander module for MD simulations [26]. AMBER is closely associated with the AMBER force field and offers a wide range of simulation options.
• CHARMM: A comprehensive program for macromolecular simulations, with a focus on structural biology [27]. CHARMM is often used for simulations of proteins, nucleic acids, and lipids, and it is closely associated with the CHARMM force field.
• OpenMM: A highly extensible and customizable toolkit for MD simulations [28]. OpenMM is designed to be platform-independent and supports a wide range of force fields and simulation algorithms. It is also well-suited for developing new simulation methods.

MD simulations are computationally intensive, requiring significant computational resources. The size and complexity of the system, the simulation time, and the desired accuracy all affect the computational cost. Simulations of large biomolecular systems, such as proteins in explicit solvent, can require days or weeks of computation on high-performance computing clusters [29]. Therefore, access to such resources is essential for many MD simulation projects. Resources such as XSEDE, PRACE, and national supercomputing centers provide researchers with access to powerful computing infrastructure and expertise.

The rapid development of graphics processing units (GPUs) has also revolutionized MD simulations. GPUs offer massive parallelism, making them well-suited for the computationally demanding tasks involved in MD simulations. Several MD software packages, including GROMACS, NAMD, and OpenMM, have been optimized to take advantage of GPUs, resulting in significant speedups compared to traditional CPU-based simulations [30].

Many thanks to our sponsor Esdebe who helped us prepare this research report.

4. Applications in Biological Research

4.1 Protein Folding and Dynamics

MD simulations have played a crucial role in understanding the complex process of protein folding. Protein folding is the process by which a polypeptide chain attains its native three-dimensional structure, which is essential for its biological function. MD simulations can provide atomistic-level insights into the folding pathways, intermediate states, and energy landscape of proteins [31]. Enhanced sampling techniques, such as REMD and metadynamics, have been used to simulate the folding of small proteins and peptides [32].

MD simulations are also used to study the dynamics of proteins, including conformational changes, domain motions, and allosteric regulation. Allostery refers to the regulation of protein function by the binding of a ligand to a site distinct from the active site. MD simulations can reveal the structural and dynamic mechanisms underlying allosteric regulation, providing insights into how proteins communicate and respond to external stimuli [33].

4.2 Enzyme Catalysis

MD simulations are increasingly used to study enzyme catalysis, providing detailed information about the reaction mechanisms, transition states, and binding energies of substrates and inhibitors. Enzyme catalysis involves a series of chemical reactions that are accelerated by the enzyme. MD simulations, in combination with quantum mechanical (QM) calculations, can provide a detailed picture of the electronic structure changes that occur during the reaction [34].

QM/MM methods combine the accuracy of QM calculations for the active site of the enzyme with the efficiency of MM calculations for the rest of the system. This approach allows researchers to study enzyme catalysis with a high level of detail while keeping the computational cost manageable. MD simulations can also be used to screen potential inhibitors of enzymes, aiding in the development of new drugs [35].

4.3 Drug Discovery

MD simulations are widely used in drug discovery to identify and optimize potential drug candidates. Structure-based drug design relies on the knowledge of the three-dimensional structure of the target protein. MD simulations can be used to dock small molecules into the binding site of the protein and to estimate the binding affinity [36]. Free energy perturbation (FEP) calculations can provide more accurate estimates of the binding affinity by calculating the free energy difference between the bound and unbound states [37].

MD simulations can also be used to study drug resistance mechanisms. Drug resistance often arises from mutations in the target protein that alter the binding affinity of the drug. MD simulations can reveal the structural and dynamic changes caused by these mutations, providing insights into how drug resistance develops and how it can be overcome [38].

4.4 Membrane Dynamics

Biological membranes are complex and dynamic structures composed of lipids and proteins. MD simulations are used to study the structure, dynamics, and function of membranes [39]. Simulations can provide insights into the organization of lipids in the membrane, the diffusion of molecules across the membrane, and the interactions between membrane proteins and lipids. Coarse-grained MD simulations, which use a simplified representation of the atoms, can be used to simulate larger membrane systems over longer timescales [40].

4.5 Nucleic Acid Structure and Function

MD simulations are used to study the structure, dynamics, and interactions of DNA and RNA. Nucleic acids are essential for storing and transmitting genetic information. MD simulations can provide insights into the flexibility of DNA and RNA, the interactions between nucleic acids and proteins, and the effects of mutations on nucleic acid structure and function [41]. Enhanced sampling techniques are used to study the conformational changes of nucleic acids, such as the transition between different DNA or RNA secondary structures [42].

Many thanks to our sponsor Esdebe who helped us prepare this research report.

5. Limitations and Challenges

Despite its power and versatility, MD simulation has some inherent limitations and challenges that must be considered when interpreting simulation results. These limitations primarily stem from the approximations involved in the force fields, the finite simulation time, and the computational cost.

5.1 Force Field Accuracy

The accuracy of MD simulations depends critically on the quality of the force field used to describe the interatomic interactions. Current force fields are based on empirical parameters that are fitted to experimental data and quantum mechanical calculations. However, these force fields are not perfect and can have limitations in accurately describing certain types of interactions, such as polarization effects, charge transfer, and non-standard residues [43].

The transferability of force fields is also a concern. Force fields are typically developed for specific classes of molecules, such as proteins or nucleic acids. Using a force field outside its intended range of applicability can lead to inaccurate results. The development of more accurate and transferable force fields is an ongoing area of research [44].

5.2 Sampling Limitations

MD simulations are limited by the finite simulation time. Biological processes often occur on timescales that are much longer than the accessible simulation time. For example, protein folding can take milliseconds or even seconds, while MD simulations are typically limited to microseconds or milliseconds. This timescale gap makes it difficult to study rare events and long-time dynamics [45].

Enhanced sampling techniques can help to overcome this limitation by accelerating the exploration of the system’s energy landscape. However, these techniques also have their own limitations and may introduce biases into the simulation results. Careful validation of the simulation results is essential to ensure that they are not artifacts of the enhanced sampling method [46].

5.3 Computational Cost

MD simulations are computationally demanding, requiring significant computational resources. The computational cost increases rapidly with the size of the system and the simulation time. Simulating large biomolecular systems, such as proteins in explicit solvent, can require days or weeks of computation on high-performance computing clusters [47].

The development of more efficient simulation algorithms and the use of GPUs can help to reduce the computational cost. However, MD simulations will likely remain computationally intensive for the foreseeable future. Access to sufficient computational resources is essential for many MD simulation projects.

5.4 Integration with Experimental Data

To obtain a more comprehensive understanding of biological systems, MD simulations should be integrated with experimental data. Experimental data can be used to validate the simulation results and to provide constraints for the simulations [48]. For example, experimental structures from X-ray crystallography or NMR spectroscopy can be used to validate the simulated structure. Experimental data on binding affinities, reaction rates, or conformational changes can be used to validate the simulated dynamics.

Integrative modeling approaches combine MD simulations with experimental data to build more accurate and reliable models of biological systems. These approaches can help to overcome the limitations of both MD simulations and experimental methods, leading to a more complete understanding of biological processes [49].

Many thanks to our sponsor Esdebe who helped us prepare this research report.

6. Future Directions

The field of MD simulation is constantly evolving, with new methods and applications being developed all the time. Some of the key future directions include:

• Development of more accurate and transferable force fields: Improving the accuracy and transferability of force fields is crucial for increasing the reliability of MD simulations. This includes developing polarizable force fields, incorporating machine learning techniques to parameterize force fields, and developing force fields that can accurately describe a wider range of chemical species [50].
• Development of more efficient simulation algorithms: Improving the efficiency of simulation algorithms is essential for simulating larger systems over longer timescales. This includes developing new integration schemes, optimizing existing algorithms for GPUs, and developing adaptive simulation methods that focus computational effort on the most important regions of the system [51].
• Integration of MD simulations with machine learning: Machine learning (ML) is becoming increasingly important in MD simulations. ML can be used to accelerate simulations, improve the accuracy of force fields, and analyze simulation data [52]. Examples include using ML to predict protein folding pathways, to identify drug binding sites, and to develop coarse-grained models of biomolecules.
• Development of multiscale modeling approaches: Multiscale modeling approaches combine MD simulations with other computational methods, such as coarse-grained simulations, continuum mechanics, and agent-based modeling, to study biological systems at different scales [53]. These approaches can provide insights into the behavior of complex systems that are not accessible to MD simulations alone.
• Application of MD simulations to new areas of biology: MD simulations are being applied to an increasingly wide range of biological problems, including personalized medicine, synthetic biology, and systems biology [54]. As computational power continues to increase and simulation methods become more sophisticated, MD simulations will play an even greater role in driving biological discoveries.

Many thanks to our sponsor Esdebe who helped us prepare this research report.

7. Conclusion

Molecular dynamics simulations have emerged as an indispensable tool in biological research, offering atomistic-level insights into complex biomolecular processes. By leveraging advancements in force fields, simulation algorithms, and computational resources, MD simulations have significantly contributed to our understanding of protein folding, enzyme catalysis, drug discovery, membrane dynamics, and nucleic acid structure-function relationships. While limitations related to force field accuracy, sampling, and computational cost remain, ongoing research efforts are actively addressing these challenges. Integrating MD simulations with experimental data and exploring innovative approaches such as machine learning and multiscale modeling will further enhance the power and applicability of MD simulations in unraveling the intricacies of biological systems. As computational capabilities continue to grow, MD simulations promise to play an increasingly central role in shaping our understanding of life at the molecular level.

Many thanks to our sponsor Esdebe who helped us prepare this research report.

References

[1] D. Frenkel and B. Smit, Understanding Molecular Simulation: From Algorithms to Applications. Academic Press, 2001.
[2] J. D. Chodera, W. C. Swope, J. W. Pitera, K. Dill, and F. Noé, “Automatic discovery of metastable states for the construction of Markov state models of macromolecular conformational dynamics,” J. Chem. Theory Comput., vol. 3, no. 1, pp. 26–41, 2007.
[3] A. Warshel, Modern Enzyme Science: Problems and Solutions. World Scientific, 2011.
[4] J. B. Anderson, “Molecular dynamics simulations of drug-receptor interactions,” Methods Mol. Biol., vol. 1268, pp. 13–26, 2015.
[5] S. J. Marrink and D. P. Tieleman, “The DLPC model for coarse-grained molecular dynamics simulations of lipid bilayers,” J. Phys. Chem. B, vol. 105, no. 36, pp. 7011–7022, 2001.
[6] T. E. Cheatham, III and M. J. Young, “Molecular dynamics simulation of nucleic acids: successes, limitations, and promise,” Annu. Rev. Phys. Chem., vol. 51, pp. 435–471, 2000.
[7] M. P. Allen and D. J. Tildesley, Computer Simulation of Liquids. Oxford University Press, 2017.
[8] T. Schlick, Molecular Modeling and Simulation: An Interdisciplinary Guide. Springer, 2010.
[9] W. L. Jorgensen, D. S. Maxwell, and J. Tirado-Rives, “Development and testing of the OPLS all-atom force field on conformational energetics and properties of organic liquids,” J. Am. Chem. Soc., vol. 118, no. 45, pp. 11225–11236, 1996.
[10] A. D. MacKerell, Jr. et al., “All-atom empirical potential for molecular modeling and dynamics studies of proteins,” J. Phys. Chem. B, vol. 102, no. 18, pp. 3586–3616, 1998.
[11] S. J. Weiner et al., “A new force field for molecular mechanical simulation of nucleic acids and proteins,” J. Am. Chem. Soc., vol. 106, no. 3, pp. 765–784, 1984.
[12] W. F. van Gunsteren et al., “GROMOS force-field parameter sets 54A7 and 54B7 for alanyl dipeptide in different solvents,” J. Phys. Chem. B, vol. 108, no. 1, pp. 177–185, 2004.
[13] P. Ren and J. W. Ponder, “Polarizable atomic multipole water model for molecular mechanics simulations,” J. Phys. Chem. B, vol. 107, no. 24, pp. 5933–5947, 2003.
[14] L. Verlet, “Computer ‘experiments’ on classical fluids. I. Thermodynamical properties of Lennard-Jones molecules,” Phys. Rev., vol. 159, no. 1, p. 98, 1967.
[15] H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, A. DiNola, and J. R. Haak, “Molecular dynamics with stochastic dynamics,” J. Chem. Phys., vol. 81, no. 8, pp. 3684–3690, 1984.
[16] S. Nosé, “A unified formulation of the constant temperature molecular dynamics methods,” J. Chem. Phys., vol. 81, no. 1, pp. 511–519, 1984.
[17] W. Shinoda, M. Shiga, and M. Mikami, “Rapid equilibration in molecular dynamics simulation by using a Nosé-Hoover chain method,” Phys. Rev. B, vol. 69, no. 13, p. 134103, 2004.
[18] A. Laio and M. Parrinello, “Escaping free-energy minima,” Proc. Natl. Acad. Sci. USA, vol. 99, no. 20, pp. 12562–12566, 2002.
[19] G. M. Torrie and J. P. Valleau, “Nonphysical sampling distributions in Monte Carlo free-energy estimation: Umbrella sampling,” J. Comput. Phys., vol. 23, no. 2, pp. 187–199, 1977.
[20] A. Laio and F. L. Gervasio, “Metadynamics: A powerful method for exploring free energy landscapes,” Rep. Prog. Phys., vol. 71, no. 12, p. 126601, 2008.
[21] D. Hamelberg, J. D. Mongan, and J. A. McCammon, “Accelerated molecular dynamics: extending the sampling range of biomolecular simulations,” J. Chem. Phys., vol. 120, no. 24, pp. 11919–11929, 2004.
[22] R. H. Swendsen and J. S. Wang, “Replica Monte Carlo for statistical mechanics,” Phys. Rev. Lett., vol. 57, no. 21, p. 2607, 1986.
[23] F. Noé, “Markov state models for an ensemble of time series,” J. Chem. Phys., vol. 128, no. 24, p. 244103, 2008.
[24] M. J. Abraham et al., “GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers,” SoftwareX, vol. 1–2, pp. 19–25, 2015.
[25] J. C. Phillips et al., “Scalable molecular dynamics with NAMD,” J. Comput. Chem., vol. 26, no. 16, pp. 1781–1802, 2005.
[26] D. A. Case et al., AMBER 2021. University of California, San Francisco, 2021.
[27] B. R. Brooks et al., “CHARMM: The biomolecular simulation program,” J. Comput. Chem., vol. 30, no. 10, pp. 1545–1614, 2009.
[28] P. Eastman et al., “OpenMM 4: A reusable, extensible, hardware independent library for high performance molecular simulation,” J. Chem. Theory Comput., vol. 9, no. 1, pp. 461–469, 2013.
[29] K. Lindorff-Larsen et al., “Improved side-chain torsion potentials for the Amber ff99SB protein force field,” Proteins, vol. 78, no. 8, pp. 1950-1958, 2010.
[30] J. E. Stone et al., “GPU-accelerated molecular modeling coming of age,” J. Mol. Graph. Model., vol. 26, no. 1, pp. 169-181, 2007.
[31] D. E. Shaw et al., “Atomic-level characterization of the structural dynamics of proteins,” Science, vol. 330, no. 6002, pp. 341-346, 2010.
[32] Y. Sugita and Y. Okamoto, “Replica-exchange molecular dynamics method for protein folding,” Chem. Phys. Lett., vol. 314, no. 1-2, pp. 141-151, 1999.
[33] N. M. Goodey and S. J. Benkovic, “Allosteric regulation and enzyme catalysis,” Nat. Chem. Biol., vol. 4, no. 1, pp. 24-33, 2008.
[34] H. Senn and W. Thiel, “QM/MM methods for biomolecular systems,” Angew. Chem. Int. Ed., vol. 48, no. 7, pp. 1198-1229, 2009.
[35] S. D. Taylor and P. J. Kastner, “Applications of QM/MM methods in drug design,” Drug Discov. Today, vol. 15, no. 13-14, pp. 525-532, 2010.
[36] M. L. Stahl and J. D. Grossfield, “Practical considerations for binding free energy calculations,” Methods Mol. Biol., vol. 924, pp. 453-490, 2012.
[37] T. P. Straatsma, “Free energy calculations: A critique,” Curr. Opin. Struct. Biol., vol. 42, pp. 1-7, 2017.
[38] P. Varma, A. Yemini, A. Scherz, Z. Selinger, and S. Fischer, “Molecular dynamics simulations for understanding drug resistance,” Drug Resist. Updat., vol. 10, no. 1-2, pp. 33-45, 2007.
[39] S. J. Marrink, H. J. Risselada, S. Yefimov, D. P. Tieleman, and A. H. de Vries, “The MARTINI force field: coarse grained model for biomolecular simulations,” J. Phys. Chem. B, vol. 111, no. 27, pp. 7812-7824, 2007.
[40] R. G. Bennett, D. P. Tieleman, and A. D. MacKerell Jr., “Molecular dynamics simulations of membranes,” Biochim. Biophys. Acta, vol. 1838, no. 1, pp. 227-240, 2014.
[41] T. E. Cheatham III and M. J. Young, “Molecular dynamics simulation of nucleic acids: successes, limitations, and promise,” Annu. Rev. Phys. Chem., vol. 51, pp. 435-471, 2000.
[42] D. H. Mathews, J. Sabina, M. Zuker, and D. H. Turner, “Expanded sequence dependence of thermodynamic parameters improves prediction of RNA secondary structure,” J. Mol. Biol., vol. 288, no. 5, pp. 911-940, 1999.
[43] C. Oostenbrink, A. Villa, A. E. Mark, and W. F. van Gunsteren, “A biomolecular force field based on the free enthalpy of hydration and solvation: The GROMOS force-field parameter sets 53A5 and 53A6,” J. Comput. Chem., vol. 25, no. 13, pp. 1656-1676, 2004.
[44] B. Roux and T. Simonson, “Implicit solvent models,” Biophys. Chem., vol. 78, no. 1-2, pp. 69-85, 1999.
[45] D. Thirumalai, C. Hyeon, N. Biyani, and R. Dima, “Principles governing the dynamics of protein folding,” Annu. Rev. Phys. Chem., vol. 57, pp. 433-464, 2006.
[46] M. Tuckerman, “Statistical mechanics: Theory and molecular simulation”. Oxford University Press, 2010.
[47] J. A. McCammon and S. C. Harvey, Dynamics of Proteins and Nucleic Acids. Cambridge University Press, 1987.
[48] A. Sali, A. V. Lomize, F. B. Malik, D. B. Karplus, and A. Šali, “Combining experimental data and modeling to determine macromolecular structures,” Structure, vol. 13, no. 7, pp. 1025-1034, 2005.
[49] S. Gräter, B. Knapp, S. R. Dsilva, R. O. Dror, D. E. Shaw, and V. Vogel, “The mechanosensitivity of talin,” J. Biol. Chem., vol. 284, no. 46, pp. 31505-31512, 2009.
[50] J. Wang, W. Wang, P. A. Kollman, and D. A. Case, “Automatic atom type and parameter derivation using quantum mechanical calculations. I. Force field implementation,” J. Comput. Chem., vol. 25, no. 9, pp. 1157-1174, 2004.
[51] B. Hess, C. Kutzner, D. van der Spoel, and E. Lindahl, “GROMACS 4: Algorithms for highly efficient, load-balanced, and scalable molecular simulation,” J. Chem. Theory Comput., vol. 4, no. 3, pp. 435-447, 2008.
[52] J. Behler, “Neural network potential-energy surfaces,” J. Phys. Chem. Lett., vol. 2, no. 21, pp. 2657-2663, 2011.
[53] W. E, J. G. Liu, and E. Vanden-Eijnden, “Heterogeneous multiscale methods: A review,” Commun. Comput. Phys., vol. 2, no. 3, pp. 367-427, 2007.
[54] J. D. Forman-Kay, R. W. Kriwacki, and H. J. Dyson, “Structural biology and the modular proteome: Intrinsically disordered proteins,” Curr. Opin. Struct. Biol., vol. 19, no. 1, pp. 14-20, 2009.