Research And Implementation Of A Cryo-EM 3D Reconstruction Algorithm

Cryo-electron microscopy is an important technique in the field of structural biology.In recent years,it has been vigorously developed because of its biomacromolecule structures with near-atomic resolution.Because the single-particle analysis can reach the atomic resolution,a lot of effective results have been obtained.The most critical problem in single-particle analysis is how to obtain the projection angle of a particle.Based on the idea of providing some candidate projection angles for each particle,a series of studies have been carried out around the two-dimensional clustering and three-dimensional reconstruction of single-particle analysis based on some machine learning algorithms.In this paper,the autoencoder neural network is used to reduce the dimension of the particle images,and then the particles are clustered in the low-dimensional space.The average images of the two-dimensional clustering are obtained quickly.The relationship between the average and the projection images which projected by the initial model is established.Based on the K-nearest neighbor algorithm,several projection images which are similar to a cluster average image are found quickly and accurately,so several candidate angles of a particle which belongs to the cluster are obtained.Through the iterations searching the real projection angle of a particle around the candidate angles and the protein three-dimensional structure are obtained.In order to acquire the performance of each module,I analyze the results of two-dimensional clustering algorithm.It is proved that the candidate angles found by KD tree algorithm is indeed distributed around the real angle.Finally,based on the two protein data of Beta-galactosidase and Trpv1,the experimental results show that the method is not only superior to the existing algorithm in the numerical of resolution and fourier shell correlation graph,but also increased by 58.3% and 35.7% respectively in speed.In addition,by analyzing the structural details,the method of this article can get a high resolution three-dimensional structure.