Related Experiment Video
Updated: Aug 6, 2026
10:25
Deep Learning-Based Segmentation of Cryo-Electron Tomograms
Published on: November 11, 2022
DeepSSETracer 2.0: Improved Deep Learning Model Performance for Protein Secondary Structure Segmentation from Cryo-EM
Bryan Hawickhorst1, Thu Nguyen1, Willy Wriggers2
1Department of Computer Science, Old Dominion University, Norfolk, Virginia, USA.
ACM-BCB ... ... : the ... ACM Conference on Bioinformatics, Computational Biology and Biomedicine. ACM Conference on Bioinformatics, Computational Biology and Biomedicine
|July 26, 2026
Summary
DeepSSETracer 2.0 significantly improves protein secondary structure segmentation in cryo-EM maps. This enhanced method boosts helix and beta-sheet detection accuracy, aiding structural biology research.
Area of Science:
- Structural Biology
- Biophysics
- Computational Biology
Background:
- Accurate segmentation of protein secondary structures (helices and beta-sheets) from cryo-EM density maps is crucial for understanding protein function.
- Existing methods face challenges with medium-resolution (5-10Å) cryo-EM data.
Purpose of the Study:
- To enhance the performance of DeepSSETracer for segmenting protein secondary structures from cryo-EM density maps.
- To investigate the impact of various technical improvements on segmentation accuracy.
Main Methods:
- DeepSSETracer, a deep learning method, was refined through modifications including normalization methods, max-pooling, activation functions, and loss calculation regions.
- Ablation studies were performed to assess the contribution of each modification.
- The performance was evaluated on 77 test cases using weighted average per-voxel F1 score.
Main Results:
- DeepSSETracer 2.0 demonstrated significant performance improvements over DeepSSETracer 1.1.
- Weighted average per-voxel F1 score increased from 62.1% to 70.3% for helix detection and from 47.8% to 62.5% for beta-sheet detection.
- Replacing batch normalization with instance normalization yielded the most substantial accuracy gains, particularly for beta-sheet detection.
Conclusions:
- Modest network tuning can substantially improve protein secondary structure segmentation in cryo-EM.
- DeepSSETracer 2.0 offers a more accurate tool for analyzing medium-resolution cryo-EM data.
- Further incremental improvements are possible within the U-Net deep learning architecture.
