Convolutional Neural Network Method for Particle Selection of Cryo-EM Micrographs

Cryo-electron microscopy(cryo-EM) has been widely used in structural biology. Molecular particle selection in cryo-EM micrographs as an important intermediate affects the quality of 3D reconstruction. A new method called RSelector is proposed with a new feature extraction network based on characteristics of cryo-EM micrographs and two sub-networks for classification and coordinate regression. Experiments on the three public cryo-EM datasets show that with less manual interference in the training stage compared with the existing common methods, the RSelector method achieves a better selection performance and also performs well in terms of resistance to label noise.