This dissertation presents a new computational framework for calculating the normal modes and interactions of proteins, macromolecular assemblies and surrounding solvents. The framework employs a combination of molecular dynamics simulation (MD) and principal component analysis (PCA). It enables the capture and visualization of the molecules' normal modes and interactions over time scales that are computationally challenging. It also provides a starting point for experimental and further computational studies of protein conformational changes.
A protein's function is sometimes linked to its conformational flexibility. Normal mode analysis (NMA) and various extensions of it have provided insights into the conformational fluctuations associated with individual protein structures. In traditional NMA, each protein requires a customized model for such analysis due to its mechanical complexity. The methodology presented here is applicable to any protein with known atomic coordinates. Because of its computational efficiency and scalability, it facilitates the study of slow protein conformational changes (on the order of milliseconds), such as protein folding.
PCA reduces the dimensionality of MD atomic trajectory data and provides a concise way to visualize, analyze, and compare the motions observed over the course of a simulation. PCA involves diagonalization of the positional covariance matrix and identification of an orthogonal set of eigenvectors or "modes" describing the direction of maximum variation in the observed conformational distribution. Consequently, slow conformational changes can be identified by projecting these dominant modes back to original trajectory data.
In this work, the new multiscale methodology was first applied to a relatively small mutant T4 phage lysozyme, establishing its equilibrium atomic thermal fluctuations and its inter-residue fluctuation correlations. These results were compared with published data obtained by NMA, by finite element methods, and by experiment. The eigenmodes captured are in quantitative agreement with previously published results. With this success on a small protein, the method was applied to the interaction of mutated hemoglobin molecules that cause sickle cell anemia and the atomic level details of which are unknown. The new methodology reveals slow motion processes of the hemoglobin-hemoglobin interaction.
MD based PCA is computationally expensive. Thus, this dissertation work also includes a widely-applicable parallel programming implementation of the modeling framework to improve its performance.
|
![[NJIT logo]](http://archives.njit.edu/vhlib/images/njit_logo.gif){kind=logo}
![[Van Houten Library logo]](images/vhl-top-image.jpg)
