Heterogeneous reconstruction algorithms for cryoEM achieve limited particle classification accuracy on real benchmark...

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Single-particle cryo-electron microscopy (cryo-EM) excels at determining static structures and complex conformational landscapes.
• Heterogeneous reconstruction algorithms can resolve thousands of maps from a single cryo-EM dataset.
• Assessing algorithm accuracy is difficult due to a lack of benchmark datasets with ground-truth labels.

Purpose of the Study:

• To develop and utilize novel benchmark datasets for rigorous assessment of cryo-EM heterogeneous reconstruction algorithms.
• To evaluate the accuracy of current heterogeneous reconstruction methods on realistic, simulated cryo-EM data.

Main Methods:

• Development of benchmark datasets using Cas9 and DNA for ground-truth per-particle structural labels.
• In silico resampling of particle stacks from these datasets.
• Challenging two popular heterogeneous reconstruction algorithms with mixed particle stacks.

Main Results:

• Existing heterogeneous reconstruction algorithms demonstrated limited accuracy in resolving encoded heterogeneity.
• Reconstruction accuracy was highly dependent on prior information for complex, multi-scale, and multi-axis heterogeneity.
• Significant errors in individual particle assignments occurred even when correct structural states were reconstructed.
• Molecular breathing motions and data collection parameters (defocus, projection angle) contributed to particle assignment errors.

Conclusions:

• Current heterogeneous reconstruction methods exhibit significant shortcomings in accuracy and particle assignment.
• Method development is needed for both cryo-EM data collection strategies and heterogeneous classification/reconstruction algorithms.
• Future work should focus on improving robustness to complex heterogeneity and data collection variations.