UNCERTAINTY QUANTIFICATION OF UNKNOWN ORIENTATIONS AND ab-initio RECONSTRUCTIONS IN SINGLE-PARTICLE Cryo-EM

Abstract

Single-particle cryo-electron microscopy (cryo-EM) is a tool which aids in the analysis of proteins and biological macromolecules (particles), to look into their molecular, biochemical and cellular processes. The biological sample is vitrified, and the particles are imaged in their near-native state, at random and unknown orientations.

At the heart of cryo-EM is attaining a three-dimensional (3D) reconstruction from two-dimensional images. Existing reconstruction methods are reviewed, beginning with classical approaches, followed by key developments, paving the way for the algorithms and methods used today. However, these existing methods produce only a single 3D structure, or at most a discrete set of 3D structures, failing to account for multiple factors of uncertainty. This thesis thus proposes a probabilistic framework. By placing distributions on the unknown orientations and 3D structure, it allows for uncertainty in the reconstruction to be modelled in a robust way. When applied to both artificial and real-world datasets, the reconstructions obtained suggest that the proposed approach is able to accurately recover the main features of the particle.