The Development And Application Of Coarse-grained Model In Protein Molecular Dynamics Simulations

Currently,computer simulation has become an important tool in scientific research.Especially,the all-atom(AA)molecular dynamics(MD)simulation has provided valuable insights into structural and functional properties of biomolecular systems.With increasing computing power and advanced sampling algorithms,current AA simulation is generally applied to the studies of biological processes,which occur at about nanometer length-scale and nanosecond timescale.However,it is a great challenge for AA simulation to study some biological phenomena beyond the above scales,like HIV-1 viral capsid assembly,cytoskeletal motions,etc.The simplified or coarse-grained(CG)model is developed to address the issue.The CG model significantly reduces the number of interaction sites by grouping a few atoms or even molecules into one site and smoothes potential energy landscapes in computer simulation.Among various CG theories,the G(?) and G(?)-like models are popular in studying protein folding dynamics.The G(?) and G(?)-like models construct CG force fields based on a single native structure of protein and hold a potential propensity to the native state on the energy profiles.Here,we develop a double-well G(?)-like model with a confining potential,to simulate the conformational switching of adenylate kinase in confinement.This CG model is parametrized on two native states of adenylate kinase,closed and open ones.In addition,a spherical confining potential,which is expressed almost same as the standard van der Waals(vdW)form,is introduced in total potential energy to mimic a confining environment.Results show that a small degree of confinement reduces the entropy of open state and stabilizes the closed state,which leads to increased energy barriers for transition.Furthermore,the analysis of transition temperatures and mean first passage times indicates that proper affinity can promote the transition from closed state to open state.This study reveals that the confining environment plays an important role in the thermodynamics and kinetics of proteins in vivo.Though the G(?) and G(?)-like models are developed for protein folding,it fails to carry out studies for the systems without a native structure.Thus,a new two-beadmultipole force field(TMFF)is proposed to more generally perform CG simulations of protein.Different from most CG models,the TMFF explicitly introduces electrostatic multipoles in the conventional two-bead CG model to improve the accuracy of electrostatic calculation.And TMFF is also practical,independent of protein's native structures or secondary-structure elements.Multiple benchmark proteins are simulated to validate the TMFF with Martini one-site CG water.The backbone root-mean-square-fluctuations(RMSFs)predicted by TMFF agree well with those by AA force field.Besides,the use of TMFF significantly saves computational cost,compared to the multipole-based AA simulation.Subsequently,various polarizable CG water models,including Martini three-site model,Cui's three-site model,Riniker's two-site model and our newly proposed two-site one,are respectively coupled with TMFF to carry out a more realistic CG simulation of protein.Tests of multiple benchmark systems show a polarizable solvent environment significantly stabilizes protein's native structures in CG simulation by TMFF and it still predicts RMSFs well.The TMFF with various water models is general,transferable and suitable for the CG studies of large biomolecules.Initial TMFF is parametrized on the statistic data of protein crystal structure library and effects of secondary-structure elements are unavoidably combined in the statistical result.To obtain statistic data with no structural propensities,the protein structure library is substituted by a protein coil library in the re-parametrizing of TMFF.The re-parametrized TMFF is used to study the protein folding.The ?s-level simulations of two small proteins,respectively helical-hairpin and beta-hairpin structures,show the re-parametrized TMFF enables protein to achieve the folding to a certain degree and guides a right path toward the folding state.The tests using the re-parametrized TMFF open a door for CG models except G(?) and G(?)-like ones to study protein folding.To further couple with the multipole-based protein force fields,a point-dipole-based one-site water model is developed based on the Martini polarizable one.The new protein force field is similar to the TMFF but largely different in the CG map and electrostatic expression.Here,the multipole-based protein force field with a point-dipole-based water model is named as point-dipole-based coarse-grained(PD-CG)force field.The PD-CG water is parametrized on the statistical data from the Martini polarizable model and AA water while the PD-CG protein is done on a protein coil library.The motions of dipoles from PD-CG water follow the theories of rigid-body rotation while the dipoles from PD-CG protein are instantaneously updated by a local frame scheme with CG beads moving.Tests of benchmark proteins show that the PD-CG force field significantly stabilizes protein's native structures,better than the newest Martini protein force field.A MD time step,10 fs is advisable in the PD-CG model.Besides,applications of PD-CG model to two large proteins reveal the PD-CG model provides a promising way to study structural properties of large biomolecules.In developing CG models,a hot issue is how to transform a CG model back to its corresponding AA representation.Here,a rigid-fragment-based and local-frame-based(RF-LF)backmapping algorithm is developed to deal with this issue.This algorithm is generic,efficient and independent of an aforehand fragment library.Besides,a charge-site-based two-bead(CBTB)CG model is proposed and parametrized on a protein coil library.The RF-LF algorithm,CBTB CG model and AA model are combined together to carry out a resolution-switching-molecular-dynamics(RSMD)simulation of protein.In the RSMD simulation,transformations between CG and AA models can be flexibly achieved at any time.The RSMD simulations of benchmark proteins indicate the RSMD significantly enhances conformational sampling around crystal structures of protein.The aforementioned discussions focus mainly on CG models in which a residue is represented by one or even more CG sites.In developing higher coarse-graining models with a few residues into one site,a great challenge about how to reasonably define a CG map appears.Here,we propose two optimization algorithms,respectively stepwise optimization imposed with the boundary-constraint(SOBC)and stepwise local iterative optimization(SLIO),to quickly and rigorously derive CG maps of large proteins.The SOBC and SLIO are respectively based on essential dynamics coarse-graining(ED-CG)and fluctuation maximization coarse-graining(FM-CG)schemes.Various proteins are tested by the existing ED-CG methods and newly proposed ones to validate these two algorithms.The derived CG models from two algorithms are simply parametrized in harmonic form by a fluctuation-matching method.The algorithms,the SOBC on ED-CG and the SLIO on FM-CG,provide a wise choice to quickly construct CG maps of large biomolecules.