Design And Implementation Of Cryo-EM Single Particle Image Restoration And 3D Reconstruction Algorithms

Due to the extremely short wavelength of electrons,the imaging resolution of the electron microscope far exceeds that of the optical microscope.This brings blessings to biologists who explore biological particles because a detailed structure means clear functions.With the rapid development of hardware and software technology,the structure of biological particles resolved by cryogenic electron microscopy represented by single particle analysis has become more and more sophisticated in recent years.However,there are still some problems with these results.This thesis aims to deal with the problem that the resolution of cryo-EM single particle three-dimensional reconstructions can still be improved from different steps and aspects.Combined with various machine learning methods,two algorithms are designed and researched to solve the problems of extremely low signal-to-noise ratio and image degradation and the problem that high resolution information cannot be improved by the three-dimensional reconstruction process itself.These two methods can be independent or cascaded together.In the cryo-EM image restoration algorithm based on deep learning,a Wiener filter is used to remove the influence of contrast transfer function first,and then noisy and noiseless paired data is prepared for powerful feature extraction and function fitting using deep autoencoder,learning a complex mapping from noisy particle images to noiseless ones.This method performed well on two sets of protein experimental data,increasing the signal-to-noise ration significantly,helped the subsequent steps of single particle analysis to improve the accuracy,and helped to increase the resolution of reconstruction.In the particle scoring algorithm based on the maximum a posteriori estimation three-dimensional reconstruction method,this article improved the linear weak phase body model in the microscope imaging process,introducing the concept of quality weight.Based on the improved imaging model,a probabilistic model is constructed and the model parameters are thereafter updated by Expectation Maximization algorithm.This method can assign Fourier pixels of particle images with different amount of contributions when inserted into three-dimensional spaces,and reconstruct a structure with higher resolution while simultaneously estimating the weight value of the pixels.The results on two sets of experimental data show that compared with other methods,the introduction of quality weight is,to a certain extent,conducive to the improvement of resolution in three-dimensional reconstruction process.