Semi-supervised Macromolecule Structural Classification in Cellular Electron Cryo-Tomograms using 3D Autoencoding Classifier

Siyuan Liu (Carnegie Mellon University), Xuefeng Du (Xi'an Jiaotong University), Rong Xi (Carnegie Mellon University), Fuya Xu (Carnegie Mellon University), Xiangrui Zeng (Carnegie Mellon University), Bo Zhou (Yale University), Min Xu (Carnegie Mellon University)

Recent advances in the Cellular Electron Cryo-Tomography (CECT) imaging technique have enabled the 3D visualization of macromolecules and other sub-cellular components in single cells in their near-native state. Automatic structural classification of macromolecules is increasingly desirable for researchers to better study and understand the features of different macromolecular complexes. However, accurate classification of macromolecular complexes is still impeded by the lack of annotated training data due to the limited expert resource for labeling full datasets. In this paper, we introduce a semi-supervised classification framework to reduce annotation burden in the macromolecule structural classification tasks. Specifically, we propose a 3D autoencoding classifier framework for simultaneous macromolecule structural reconstruction and classification. Our framework jointly optimizes two branches of network using both labeled and unlabeled data during training phase. Extensive experiments demonstrate the effectiveness of our approach against other semi-supervised classification approaches on both real and simulated datasets. Our approach also achieves competitive results in terms of macromolecule reconstruction. To our best knowledge, this is the first work to address the task of semi-supervised macromolecule structural classification in CECT.


Paper (PDF)

title={Semi-supervised Macromolecule Structural Classification in Cellular Electron Cryo-Tomograms using 3D Autoencoding Classifier},
author={Siyuan Liu and Xuefeng Du and Rong Xi and Fuya Xu and Xiangrui Zeng and Bo Zhou and Min Xu},
booktitle={Proceedings of the British Machine Vision Conference (BMVC)},
publisher={BMVA Press},
editor={Kirill Sidorov and Yulia Hicks},