Enhancing Density Maps by Removing the Majority of Particles in Single Particle Cryogenic Electron Microscopy Final Stacks

Cryogenic electron microscopy is essential for determining near-atomic structures of biological macromolecules. Despite its reliance on averaging numerous low-signal-to-noise images, the optimum number of particles needed for specific resolution remains unknown. This limitation hinders progress in sample analysis and preparation methods.

To tackle this, we introduce an iterative sorting method, CryoSieve. Standard protocol selection includes two-dimensional and three-dimensional classification, other protocol sorting criteria such as the normalized cross-correlation method, the angular graph consistency approach, and the non-alignment classification are currently in use. Extensive experiments demonstrate that CryoSieve outperforms other cryo-EM particle sorting algorithms, revealing that most particles are unnecessary in final stacks.

The minority of particles remaining in the final stacks yields very high-resolution amplitude in reconstructive density maps. For some data sets, the size of the finest subset approaches the theoretical limit. In cryo-EM, sample preparation hinders workflow.

Due to the lack of standard metrics for protocol comparison, the ratio of selected to collected particles could serve as a quality metric. Examining their spatial and temporal distribution may also highlight key physical factors in preparation effectiveness.