Content area
Full text
Introduction
Recent advancements in cryogenic electron microscopy (cryo-EM) hardware and image processing software have ushered in a transformative era in structure determination, establishing it as the predominant method in structural biology. Despite these advances, a significant challenge that continues to impede structural analysis is the issue of preferred orientation1, 2, 3, 4, 5, 6–7. Ideally, biological macromolecules of interest should exhibit uniformly random orientations within the vitreous amorphous ice. However, it is commonly observed that samples tend to adopt a specific, preferred orientation due to interactions at the air-water or the support-water interface8,9. Using conventional computational methods to analyze preferred orientation data often results in significant artifacts in density maps10,11.
Numerous attempts have been made to address the preferred orientation in cryo-EM, focusing mainly on grid preparation and data collection. Techniques such as using detergents12, 13, 14–15, ice thickening16, shortening the spot-to-plunge time17, and biomolecule modifications18 have shown promise with specific proteins but often require time-consuming and costly condition screening. Replacing the grid and foil, as well as introducing graphene supports, has been explored to equalize particle pose distribution19,20, but their success is case-dependent or only partially effective. Tilt collection strategy4 offers an alternative by bypassing sample preparation challenges, but it introduces drawbacks such as reduced image acquisition efficiency, increased beam-induced movement, elevated noise levels due to the longer path electrons must travel, and the need for precise defocus gradient estimation8. Although per-particle CTF and motion refinements have been proposed to mitigate resolution drops from tilt collection21,22, these methods involve complex parameter adjustments and can be unstable, particularly for small proteins, which are more likely to suffer from preferred orientation than larger ones. Despite these efforts, tilt collection remains one of the most effective solutions. On the other hand, the computational aspect remains underexplored, particularly concerning whether advanced algorithms can be developed to reconstruct high-resolution density maps from previously unsolved preferred orientation datasets.
When a dataset exhibits a preferred orientation problem, particles from non-preferred views are typically present, but in much lower quantities compared to those from preferred views. Achieving isotropic reconstruction depends heavily on the number of particles captured from these non-preferred orientations. If...