Automated Segmentation and Correction of Missing-Wedge Artifacts in Cryo-Electron Tomography Maps by Shape-Constrained Deconvolution

Abstract

Cryo-electron tomography maps often exhibit considerable noise and anisotropic resolution, due to the low-dose requirements and the missing wedge in Fourier space. These spurious features are visually unappealing and, more importantly, prevent an automated segmentation of geometric shapes, often requiring a highly subjective, labor-intensive manual tracing. We continue the development of a computational strategy for objectively denoising and correcting missing-wedge artifacts in the case of repetitive basic shapes, such as filamentous structures, membranes, and other sub-tomogram features. In this approach, we use a template and a non-negative “location map” to constrain the deconvolution scheme, allowing us to recover, to a considerable degree, the information lost in the missing wedge. Whereas a previous implementation using constrained optimization showed the merit of the approach, the numerical methods were time-consuming and relied on proprietary libraries, preventing a wider adoption of our methods. Here, we explore a novel implementation based on iterative Richardson Lucy Deconvolution that can be GPU-accelerated and freely disseminated. We apply our method to cellular tomograms from collaborating laboratories and find a better overlap with the experimental map than manual tracing of cytoskeletal filaments was able to achieve. In addition, we demonstrate that our method can also be used for membrane detection and for the detection of sub-tomogram features.