Author: Swathi Adinarayanan

Publisher:

Published: 2021

Total Pages:

ISBN-13:

"The three-dimensional (3D) structures of biological complexes obtained from Single-Particle Cryo-Electron Microscopy (SPA-CryoEM) provide crucial structural insights that help in the determination of their functions. One of the most challenging aspects of SPA-CryoEM is sample heterogeneity. In my research project, I address two different issues found in the processing of highly heterogeneous samples. My first aim is focused on obtaining the best possible reconstruction of the predominant conformation while my second pursues to obtain reconstructions of all different conformations in the heterogeneous sample. Current approaches fail to address these two problems in highly heterogeneous datasets. Hence, I developed two novel computational methods to address these two challenges separately: Directional Pruning and Directional RANSAC. Directional Pruning is a method that enables us to remove remanent heterogeneities or the artifacts in the difficult datasets which are not removed by previous screening steps. This method uses directional classes, which groups particles with similar orientation together and performs 2D classification on these groups, classifying particles into two classes per direction. The particles belonging to the least populated 2D classes are removed from every directional group. The retained particles are considered homogenous particles. My hypothesis is that the retained homogenous particles will improve the quality of the final reconstruction. To test the hypothesis, different datasets of structurally heterogeneous samples were processed with directional pruning. The results showed only a slight improvement but did not really improve the quality (resolution) of the reconstruction. In order to understand my results further, I developed a new set of experiments. These experiments were designed to test the theory of the maximum likelihood (ML) algorithm used in processing of the datasets. Our results show that when I provide a good initial volume, ML algorithm automatically disregards remanent heterogeneities/artifacts present in the dataset without requiring any screening or pruning process. Directional RANSAC is a method that provides structural information of many different conformations in heterogeneous datasets. Directional classes are used to generate a number of class averages in all angular directions. Then, class averages are randomly selected along with their angular information to generate multiple 3D maps. Principal Component Analysis (PCA) is performed on generated 3D maps to identify the underlying structural components. These structural components are clustered and averaged on the basis of their similarity. Then, multiple 3D maps, representing different conformations of the macromolecule in the heterogeneous dataset are produced. The proposed approaches were tested with different structural heterogeneous datasets and the results are found to be satisfactory in comparison with current approaches.Moreover, these 3D maps can be used as reference volume in further 3D classification/3D refinement steps to obtain final 3D reconstructions of the macromolecule of interests in many different conformations"--