Methodology Studies On Resolution Measurement Of Three-dimensional Cryo-electron Microscopy Image Reconstructions

Three-dimensional structure information of biological macromolecules is crucial for understanding life activities and solving medical problems.Electron cryo-microscopy(cryo-EM)is an important method to obtain the 3D structures of biological macromolecules in structural biology.The objective quality validation method of 3D reconstruction is an important factor for stimulating the development of 3D cryo-EM.As an effective indicator of quality validation of 3D reconstruction,resolution measurement is of great importance to cryo-EM.In this paper,we have done the following studies:We proposed a new protocol to measure the resolution from a single reconstructed 3D density map.This approach estimates the spectral signal-to-noise ratio(SSNR)of a 3D reconstructed map by computing the ratio of signal power to noise power in frequency domain.The power distributions of signal and noise are estimated from structure particle region and its surrounding region segmented by applying a mask procedure.The proposed protocol of calculating SSNR,which we term as mask-SSNR(mSSNR),providing a direct calculation of SSNR based on its original definition.The mSSNR neither splits the dataset into halves to reconstruct two 3D maps like the FSC approach,nor requires any original images or intermediate data like other existing SSNR calculation methods in this field.We analyzed the changes of noise pattern in the process of reconstruction.The noise power can be approximately considered to be uniformly random distributed in 2D particle images.However,after 3D reconstruction process,the noise power changes with the positions of different distance to the center in 3D density map.We first used a sphere mask to get rid of the corner regions,and then introduced a fitting parameter to make the estimation of inside-mask noise power,which was calculated from outside-mask noise power,more accurate.We further analyzed the effects of mask size and mask type on the calculation of spectral signal-to-noise ratio and used SSNR to measure the resolution of 3D reconstruction based on relevant threshold criteria.We performed experiments on both the synthetic and real-world datasets.By comparing the resolutions calculated from several resolution measurement methods and relevant threshold criteria,we show that mask-SSNR is a simpler and more direct estimation of SSNR than the existing estimation approaches.