Molecular dynamics simulations: Parameter evaluation, application and development

Molecular dynamics (MD) simulation is a theoretical technique for investigating the physical properties of a wide variety of molecules. This dissertation contains my studies on three important parts of the MD simulation: evaluation of parameters in empirical energy functions widely used in MD simulations, application of MD simulation on experimentally interested biological molecules and development of new methods for constraint dynamics simulations. All the work in this thesis made use of CHARMM as an MD simulation tool.; The MD simulation uses empirical energy functions parameterized by a set of parameters. These parameters play an important role in the quality of the simulations. I evaluated nine parameter sets from Harvard University and Molecular Simulations, Inc. for protein simulations by the MD simulations of hydrated form of carboxy-myoglobin and interleukin-1{dollar}beta,{dollar} which are rich in two typical protein structure motifs, helix and {dollar}beta{dollar} sheet structures respectively. It is found that some sets are good at representing helical structure proteins while others are good at {dollar}beta{dollar} sheet proteins. But all of them need improvement on representing motions at low temperature.; Experimental evidence indicates that the 1A coiled-coil domains of the Intermediate Filament (IF) proteins consisting of coiled human keratins 1 and 10 (K1 and K10) are "hot spots" for substitutional mutations. Some of these mutations are correlated to the human skin diseases--epidermolytic hyperkeratiosis (EH) and epidermolysis bullosa simplex (EBS). The MD simulation technique is used here for the first time to model and simulate these proteins to elucidate the molecular-level effects of these mutations. Lacking the experimental crystal structures, the initial structure of 1A domain of the wild type Intermediate Filament protein and its mutants were modeled from scratch to reproduce the well-known properties of the proteins of this kind followed by identical MD simulations. The important result is that the simulation data shows a clear structure difference between mutants and the wild type.; Because of the presence of very light atoms such as hydrogens and strong bonding forces in molecular systems, there exist high frequency motions such as bond stretching and bond-angle bending. These highly oscillatory components hinder any explicit integrator from using longer time steps during the molecular dynamics simulations. The SHAKE algorithm is commonly used to constrain bond lengths by applying holonomic distance constraints. While this approach has been successful on bond stretching, it fails on bond-angle bending because of introduced rigidity. To combine both the benefits of free dynamics and the constraint method, we propose two new constraint algorithms to constrain the local geometry at energy minimum distances at each dynamics step, instead of using ideal equilibrium bond lengths throughout a simulation. The algorithms were tested on simulations of collisions of two diatomic molecules and two triatomic molecules (water), as well as on a box of water with periodic boundary conditions. In all cases, our algorithms gave results that are closer to free dynamics than the SHAKE algorithm did.