Massively Parallel Implementation of Steered Molecular Dynamics in Tinker-HP: Comparisons of Polarizable and Non-Polarizable Simulations of Realistic Systems

Abstract

Steered Molecular Dynamic (SMD) is a powerful technique able to accelerate rare events sampling in Molecular Dynamics (MD) simulations by applying an external force to a set of chosen atoms. Despite generating non-equilibrium simulations, SMD remains capable of reconstructing equilibrium properties such as the Potential of Mean Force (PMF). Of course, one would like to use all types of force fields (FF) ranging from classical ones to more advanced polarizable models using point induced dipoles and distributed multipoles such as AMOEBA. To enable such studies, the SMD methodology has been implemented in the framework of the massively parallel Tinker-HP software allowing for both long polarizable and non-polarizable MD simulations of large proteins. To validate this new implementation, we first compared the Tinker-HP SMD results to the Literature. Tests have been performed on three different benchmark systems: the M-A deca-alanine (112 atoms), the ubiquitin (9737 atoms) and the CD2CD58 complex (97594 atoms). Non-polarizable ( AMBER99, AMBER99SB, CHARMM22/CMAP and OPLS-AA/L) and polarizable (AMOEBAPRO13 and AMOEBABIO18) force fields have been used. For each one of them, PMFs have been reconstructed and compared in terms of free energy barrier and hydrogen bonding fluctuations behaviour over time.Using a SMD velocity of 0.01 Å/ps applied to a set of 20 trajectories, we show that polarizable and non-polarizable force fields do not always agree. As it could be anticipated, strong discrepancies are noticed between polarizable and non-polarizable models when considered in vacuum, whereas results are more comparable when a water environment is added. However, for the largest system, i.e. the CD2CD58 complex, strong differences related to the modeling of a salt bridge are noticed exhibiting some potential issues with classical FFs. Overall, such simulations highlight the importance of the inclusion of polarization effects as PMFs free energy barriers computed with AMOEBA always decrease compared to non-polarizable force fields.

Polarizable embedding QM/MM: the future gold standard for complex (bio)systems?

“…Therefore coupling these methods with polarizable multiscale models is generally straightforward, and only requires the MD engine to support the specific sampling method that one wants to apply. The ingredients needed for the simplest methods (such as harmonic potentials to perform US) are already implemented in almost all MD software, including Tinker, 135 and can be easily exploited to perform free energy calculations. In order to use more complex and flexible CV based methods, it is very convenient to exploit specific codes that provide a general interface to do so.…”

Accurate Biomolecular Simulations Account for Electronic Polarization

“…The description of hydrogen bonds in standard force fields, however, does not contain important contributions, e.g., from polarization and partially covalent character (Babin et al, 2006). It was demonstrated in many cases including structure of water (Dang, 1998), binding of ligands to proteins (Friesner, 2005; Jiao et al, 2008), and protein folding and unfolding (Morozov et al, 2006; Freddolino et al, 2010; Piana et al, 2011, 2014; Huang and MacKerell, 2014; Lemkul et al, 2016; Célerse et al, 2019) that polarizability contributes significantly to the accuracy of simulations of structures with hydrogen bonds. Also, salt bridging between amino acids is likely overestimated in strength when the effects of polarization are not included (Friesner, 2005; Vazdar et al, 2013; Debiec et al, 2014; Ahmed et al, 2018; Célerse et al, 2019; Mason et al, 2019).…”

“…It was demonstrated in many cases including structure of water (Dang, 1998), binding of ligands to proteins (Friesner, 2005; Jiao et al, 2008), and protein folding and unfolding (Morozov et al, 2006; Freddolino et al, 2010; Piana et al, 2011, 2014; Huang and MacKerell, 2014; Lemkul et al, 2016; Célerse et al, 2019) that polarizability contributes significantly to the accuracy of simulations of structures with hydrogen bonds. Also, salt bridging between amino acids is likely overestimated in strength when the effects of polarization are not included (Friesner, 2005; Vazdar et al, 2013; Debiec et al, 2014; Ahmed et al, 2018; Célerse et al, 2019; Mason et al, 2019). For instance, the interaction of acidic side chains of glutamate and aspartate with cations is overestimated in strength in classical non-polarizable force fields (Patel et al, 2009; Duboué-Dijon et al, 2018a), while treatment of polarizability in solvent relaxation affects salt bridge dissociation (Célerse et al, 2019).…”

High-Resolution Mining of SARS-CoV-2 Main Protease Conformational Space: Supercomputer-Driven Unsupervised Adaptive Sampling

“…Our group has been involved for many years in the demonstration of the importance of considering explicit manybody effects in classical MD and free energy methods through the use of polarizable force fields (PFFs). [14][15][16][17][18] Indeed, polarization affects solvation and modify the stability of secondary and quaternary structures of proteins, playing therefore a crucial role in the definition of the conformational space of a protein. Applying such methods to COVID-19 research could provide additional insights for drug modelers and experimental teams.…”