Difference between revisions of "2019Li Deep"
(Created page with "== Citation == Li, X.; Lin, Y.; Liu, Q.; McSweeney, S. & Yoo, S. Picking Particles in Cryo-EM Micrographs without Knowing the Particle Size 2019 New York Scientific Data Sum...")
Latest revision as of 22:02, 14 January 2020
Li, X.; Lin, Y.; Liu, Q.; McSweeney, S. & Yoo, S. Picking Particles in Cryo-EM Micrographs without Knowing the Particle Size 2019 New York Scientific Data Summit (NYSDS), 2019 , 1-8
Picking particles in cryo-electron microscopy images, or micrographs, is a crucial first step in reconstruction of high resolution 3D structures. In this paper, motivated by a state-of-the-art object detection model, we proposed a deep learning based automatic particle pick model. Existing models usually fail to pick particles on a dataset (target domain) if they were trained on another dataset (source domain) when the particle sizes are significant different between source and target domains. We proposed diverse size data augmentation to solve this problem. Furthermore, we use the prior knowledge that the particle sizes should be similar within one micrograph as an additional loss. Once trained, the proposed model can pick particles without knowledge of the particle size. Compared with two state-of-the-art deep learning based particle picking models, our proposed model significantly outperformed on cross domain settings, while comparable on single domain settings. Furthermore, the proposed model is much faster than comparison models.