Reference-based Three-dimensional Classification Of Cryo-EM Single-particle Imaging Data

In recent years,due to the development of Cryo-EM technology,high-resolution structures using Cryo-EM have emerged in the field of structural biology.Breakthroughs in Cryo-EM hardware devices such as direct electron detector technology,software algorithms such as maximum likelihood estimation,and the application of Bayesian ideas are key.The Cryo-EM has the advantages of relatively simple sample preparation and reduced sample damage.Physiological conditions that the samples are maintained in aqueous environment during measurement.Cryo-EM provides a clear structure and contains molecular dynamics information and has become a popular method for structural biology detection following the methods of nuclear magnetic resonance and X-ray diffraction.A number of Cryo-EM image processing toolkits have appeared in the field of Cryo-EM,some of which cover the complete calculation process of Cryo-EM image processing,and can be processed in a short time with the support of high-performance computers As a result,it is very important to study algorithms for improving the accuracy of reconstruction.However,since the sample is present in the aqueous solution before freezing,the conformation of the sample in the solution is not uniform.This has caused problems in the reconstruction of Cryo-EM imaging data.Thermal dynamics of the sample in aqueous solution will result in low resolution of the moving part after reconstruction.Due to the development of computer technology,MD simulation has made great progress in both time scale and accuracy.Therefore,this paper proposes a feasible reference 3D classification based on molecular dynamics analysis.It is intended to obtain a priori information of the sample in a dynamic simulation manner to obtain a stable structure that may exist in the sample,and to help distinguish the multi-conformation changes present in the Cryo-EM image.The classification method uses expectation maximization to achieve maximum likelihood estimation,and describes the image angle parameters by obtaining the continuous probability density of each image at the spatial angle.This paper uses the elastic network model and molecular perturbation model to generate a series of structural change models.Then using the generated models to obtain projection images with different signal-to-noise ratios and different spatial angles.The classification effect of the reference-based classification method under different influencing factors was verified by numerical simulation.It is found that in the system with relatively small conformational changes,the referencebased approach has more advantages.