Segmentation and Structure Determination in Electron Microscopy

One of the goals of biology is to be able to understand the structure and interaction of macromolecules, to be able to better understand life at a macromolecular level. One of the most important inventions that revolutionized the study of macromolecular structures is that of the electron microscope. Electron microscopes are used for studying three dimensional structures of macromolecular assemblies using 2D and 3D geometry. The underlying principle of 3D reconstruction from 2D projections is well understood and forms the basis of electron microscopy. Depending on the type of structure under investigation, either electron tomography is used where the structures are heterogeneous or in case that they are homogeneous single particle electron microscopy is used. Whatever be the underlying source of the data, tomography or single particle, they involve significant amounts of computational problems. Many of these problems have been studied in other branches of computer science, like in computer vision and machine learning. However, until very recently, there has not been a significant exchange of ideas between these two disparate communities. This work is a step in that direction. We study two problems: the first related to the well-studied problem of segmentation but in the context of electron tomography. The second relates to that of studying the macromolecular structure of actin-myosin interaction using 3D reconstruction of single particle electron microscopic data. We hope that this would be the beginning of a formal interaction between the two fields that has the potential to enrich each other tremendously.