Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SemiETPicker: Fast and Label-Efficient Particle Picking for CryoET Tomography Using Semi-Supervised Learning

Linhan Wang, Jianwen Dou, Wang Li, Shengkun Wang, Zhiwu Xie, Chang-Tien Lu, Yinlin Chen

TL;DR

This paper tackles the bottleneck of particle picking in CryoET by proposing SemiETPicker, a fast, label-efficient framework that fuses a heatmap-based end-to-end detector with a teacher–student semi-supervised learning loop. The method introduces multi-view pseudo labeling and a CryoET-specific DropBlock augmentation to improve reliability under sparse labeling, with an EMA-based teacher guiding the student on unlabeled data. On the CZII dataset, SemiETPicker yields substantial gains over supervised baselines and outperforms prior SSL approaches, while maintaining high efficiency through a simple max-pooling postprocessing step. Overall, the work demonstrates that leveraging unlabeled CryoET data with carefully designed SSL components can significantly enhance high-resolution protein localization within densely packed tomograms, accelerating structural biology analyses in cellular contexts.

Abstract

Cryogenic Electron Tomography (CryoET) combined with sub-volume averaging (SVA) is the only imaging modality capable of resolving protein structures inside cells at molecular resolution. Particle picking, the task of localizing and classifying target proteins in 3D CryoET volumes, remains the main bottleneck. Due to the reliance on time-consuming manual labels, the vast reserve of unlabeled tomograms remains underutilized. In this work, we present a fast, label-efficient semi-supervised framework that exploits this untapped data. Our framework consists of two components: (i) an end-to-end heatmap-supervised detection model inspired by keypoint detection, and (ii) a teacher-student co-training mechanism that enhances performance under sparse labeling conditions. Furthermore, we introduce multi-view pseudo-labeling and a CryoET-specific DropBlock augmentation strategy to further boost performance. Extensive evaluations on the large-scale CZII dataset show that our approach improves F1 by 10% over supervised baselines, underscoring the promise of semi-supervised learning for leveraging unlabeled CryoET data.

SemiETPicker: Fast and Label-Efficient Particle Picking for CryoET Tomography Using Semi-Supervised Learning

TL;DR

This paper tackles the bottleneck of particle picking in CryoET by proposing SemiETPicker, a fast, label-efficient framework that fuses a heatmap-based end-to-end detector with a teacher–student semi-supervised learning loop. The method introduces multi-view pseudo labeling and a CryoET-specific DropBlock augmentation to improve reliability under sparse labeling, with an EMA-based teacher guiding the student on unlabeled data. On the CZII dataset, SemiETPicker yields substantial gains over supervised baselines and outperforms prior SSL approaches, while maintaining high efficiency through a simple max-pooling postprocessing step. Overall, the work demonstrates that leveraging unlabeled CryoET data with carefully designed SSL components can significantly enhance high-resolution protein localization within densely packed tomograms, accelerating structural biology analyses in cellular contexts.

Abstract

Cryogenic Electron Tomography (CryoET) combined with sub-volume averaging (SVA) is the only imaging modality capable of resolving protein structures inside cells at molecular resolution. Particle picking, the task of localizing and classifying target proteins in 3D CryoET volumes, remains the main bottleneck. Due to the reliance on time-consuming manual labels, the vast reserve of unlabeled tomograms remains underutilized. In this work, we present a fast, label-efficient semi-supervised framework that exploits this untapped data. Our framework consists of two components: (i) an end-to-end heatmap-supervised detection model inspired by keypoint detection, and (ii) a teacher-student co-training mechanism that enhances performance under sparse labeling conditions. Furthermore, we introduce multi-view pseudo-labeling and a CryoET-specific DropBlock augmentation strategy to further boost performance. Extensive evaluations on the large-scale CZII dataset show that our approach improves F1 by 10% over supervised baselines, underscoring the promise of semi-supervised learning for leveraging unlabeled CryoET data.
Paper Structure (14 sections, 2 figures, 4 tables)

This paper contains 14 sections, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Problem Description
  4. An End-to-End Object Detector for Particle-Picking
  5. Teacher-Student Co-training and Multi-View Pseudo Labeling
  6. CryoET-specific DropBlock Augmentation
  7. Experiment
  8. Dataset
  9. Implementation Details
  10. Comparison with Other Methods
  11. Ablation Studies and Efficiency Analysis
  12. Conclusion
  13. Compliance with ethical standards
  14. Acknowledgments

Figures (2)

  • Figure 1: Overview of the SemiETPicker pipeline. The student model is jointly trained on labeled and unlabeled data using a combination of reweighted MSE loss and consistency loss. To reduce pseudo label uncertainty, we generate multi-view samples of unlabeled images via multiple flipping for the teacher model. The teacher is updated as the exponential moving average (EMA) of the student, enabling mutual improvement through a co-training loop.
  • Figure 2: Left: CutOut augmentation. Right: CryoET-specific DropBlock augmentation.