An Automatic Particle Picking Algorithm Based On Deep Learning In Cryo-EM

Nowadays,it becomes very common to solve the 3D structure of protein based on single particle electron cryo-microscopy(cryo-EM)in the filed of structural biology.Throughout the whole process,the particle picking process costs a lot of effort and time.During the particle picking process,people firstly label some sample particles,then picking the particle based on the semi-automatic particle selection algorithm.At lastl,manually removing some false-positive particles such as ice and carbon film.So it is an urgent need for a new automated particle picking algorithm to reduce the participation of people,which can also be used to remove some false positive particles to ensure the accuracy of the selection of particles.In this context,a fully automated particle picking algorithm called DeepPicker is designed and implemented by image processing technology.DeepPicker is designed based on transfer learning and employs a novel cross-molecule training strategy to capture common features of particles from previously-analyzed micrographs,and thus does not require any human intervention during particle picking.Besides,DeepPicker takes use of Canny edge detection algorithm and the single-connected domain analysis to eliminate the false particles,such as ice and carbon film,which greatly reduces the human participation and achieves a fully automated particle picking process..Finally,tests on the recently-published cryo-EM data of three complexes have demonstrated that our algorithm can successfully accomplish the human-level particle picking process and identify a sufficient number of particles that are comparable to those picked manually by human experts.