Developing protein force fields for implicit solvent simulation

Molecular dynamics simulation is widely used in research of biomolecule properties and biomolecular processes, such as protein-protein interactions, protein ab initio folding and protein domain-domain interactions with a linker. Previous studies have shown that accuracy and efficiency of such simula...

Full description

Saved in:
Bibliographic Details
Main Author: Zhang, Haiping
Other Authors: Lu Lanyuan
Format: Theses and Dissertations
Language:English
Published: 2018
Subjects:
Online Access:http://hdl.handle.net/10356/74803
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:Molecular dynamics simulation is widely used in research of biomolecule properties and biomolecular processes, such as protein-protein interactions, protein ab initio folding and protein domain-domain interactions with a linker. Previous studies have shown that accuracy and efficiency of such simulations depends heavily on whether the force field is incorporated with explicit solvent. Although computational resources have increased tremendously in recent years, for large systems with explicit water, it is still prohibitive to achieve a simulation time scale between microseconds and milliseconds for many labs, which is the minimal time for most biochemical processes to complete. In contrast, implicit solvent models can significantly increase computational efficiency by reducing the total degrees of freedom in simulations. However, current Generalized Born models (GB) or GB with solvent accessible surface area based (GBSA-based) force fields are usually less accurate than the explicit solvent counterparts and still require improvement. In my study, I developed two methods to improve solvent-free force fields. One method that I developed is to improve the GB-Neck2 model combined with ff14SBonlysc force field, which is one of the most accurate implicit solvent models in literature. I implemented a cMAP potential energy term to adjust the secondary structure propensity for each type of residue. In addition, non-bonded parameters whose side chains contain aromatic rings were modified to better mimic pi-pi interactions. Results from simulations using my method show significantly improved performance in predicting secondary structure propensity as well as tertiary structure accuracy. A test set of 19 small peptides with experimentally determined native PDB structures and with diversified secondary structures was constructed to test the accuracy of my force field. In 16 (84.2%) of these cases, the native structures were correctly produced from extended starting structures by replica-exchange molecular dynamics simulations, using a RMSD criterion of 0.45 nm. This result is significantly superior to that using the original force field with the success ratio 47.3 % (9 in 19 cases). Additionally, some small proteins were selected to investigate the force field’s performance on larger systems, and better results from my modified parameters were also observed for those cases. Furthermore, the free energy surfaces from the protein simulations illustrate that my force field produces global free energy minima in the vicinities of native structures, while a broad range of conformations can still be sufficiently sampled in the simulations. Hence, I developed a new GB-based atomistic force field with improved ability to produce secondary and tertiary structure. My force field can efficiently sample the conformational space of peptides and small to medium sized proteins. The other method is to develop a new force field by incorporating a distance dependent dielectric constant, a pairwise statistical potential and a modified dihedral energy correction (cMAP) term together, to achieve MD simulations of proteins with high practical speed and acceptable accuracy. The pairwise statistical potential has been widely used as score term in protein-protein docking, while in our study, it shows potential application in molecular dynamics simulation as well. This force field adopts bonded parameters directly from the GROMOS54a7 force field, while adding a newly tuned cMAP to bias the backbone phi-psi distribution to the phi-psi distribution from the Protein Coil Library. The cMAPs of amino acid residues were further manually modified in order to achieve better performance for the training set. The modified force field is able to fold peptides ab initio with reasonable alpha helix/beta sheet propensity, maintaining the protein’s secondary and tertiary structure. In Chapter 4, metadynamics simulation approaches were used to explore the conformational dynamics of the dengue virus envelope (E) protein. The E protein undergoes large-scale conformational changes during the viral life cycle, including acidic pH induced changes in the endosome that enable fusion and subsequent infection. Antibodies which block such changes have the potential to inhibit viral infection. The simulations demonstrate the potential for such enhanced sampling approaches to successfully predict differences in the protein complex free energy landscape in response to changes in pH, antibody binding, protein mutations etc., and thus show promise as a tool in future rational therapeutics development. All the three projects in the thesis are related to enhanced sampling methods in molecular dynamics simulations. The implementation of implicit solvent models is a common strategy to achieve better conformational sampling in simulations, and the first two projects were designed for the force field developments in solvent-free simulations. In the last project, the Metadynamics approach, which is one of the most popular enhanced sampling methods, was adopted to explore the protein conformational changes in the dengue virus.