A deep learning approach to the automatic detection of alignment errors in cryo-electron tomographic reconstructions

• DOI: 10.1016/j.jsb.2023.108056
• Publication Date: 2023-12-14T03:41:34Z
• Authors: F.P. de Isidro-Gómez; J.L. Vilas; P. Losana; J.M. Carazo; C.O.S. Sorzano
• Keywords: Fiducial marker; Tomographic reconstruction

Electron tomography is an imaging technique that allows for the elucidation of three-dimensional structural information biological specimens in a very general context, including cellular situ observations. The approach starts by collecting set images at different projection directions tilting specimen stage inside microscope. Therefore, crucial preliminary step to precisely define acquisition geometry aligning all tilt common reference. Errors introduced this will lead appearance artifacts tomographic reconstruction, rendering them unsuitable sample study. Focusing on fiducial-based strategies, work proposes deep-learning algorithm detect misalignment reconstructions analyzing characteristics these fiducial markers tomogram. In addition, we propose designed tomogram with which feed classification case alignment does not provide location markers. This open-source software available as part Xmipp package Scipion framework, and also through command-line standalone version Xmipp.