Disentangling semantic features of macromolecules in Cryo-Electron Tomography

TL;DR

This paper introduces a 3D Variational Autoencoder that effectively disentangles structure, orientation, and shift features of macromolecules in Cryo-ET images, improving recognition and analysis.

Contribution

It presents a novel 3D Spatial Variational Autoencoder that explicitly separates key semantic features of macromolecules in Cryo-ET data, aiding downstream analysis.

Findings

Effective disentanglement of structure, orientation, and shift features.

Improved recognition accuracy on synthesized and real datasets.

Robust performance across different domains.

Abstract

Cryo-electron tomography (Cryo-ET) is a 3D imaging technique that enables the systemic study of shape, abundance, and distribution of macromolecular structures in single cells in near-atomic resolution. However, the systematic and efficient $\textit{de novo}$ recognition and recovery of macromolecular structures captured by Cryo-ET are very challenging due to the structural complexity and imaging limits. Even macromolecules with identical structures have various appearances due to different orientations and imaging limits, such as noise and the missing wedge effect. Explicitly disentangling the semantic features of macromolecules is crucial for performing several downstream analyses on the macromolecules. This paper has addressed the problem by proposing a 3D Spatial Variational Autoencoder that explicitly disentangle the structure, orientation, and shift of macromolecules. Extensive experiments on both synthesized and real cryo-ET datasets and cross-domain evaluations demonstrate the efficacy of our method.