Frame Alignment

The enabling acquisition strategy where total electron exposure is divided into multiple short sub-frames rather than a single long exposure. This temporal sampling allows computational tracking of specimen movement. Typical dose rates range from 5–20 e⁻/Ų/s, with individual frame exposures of 0.05–0.2 seconds. Without fractionation, motion blur would irreversibly degrade high-resolution information. The direct electron detector's fast readout speed makes this possible, capturing 20–60 frames per movie to create a time-resolved record of the sample's trajectory under the beam.

Beam-induced motion occurs at multiple spatial scales requiring hierarchical correction. Global motion involves rigid translation and rotation of the entire field of view due to stage drift and bulk sample warping. Local motion describes non-uniform, spatially variant deformations caused by doming, buckling, and anisotropic contraction of the vitreous ice layer. Modern algorithms like MotionCor2 apply a time-varying deformation grid to model these complex per-patch trajectories, correcting movements as small as 0.1 Å to preserve high-resolution signal that would otherwise be lost to blurring.

A critical computational compensation applied during alignment that accounts for cumulative radiation damage. As the electron beam interacts with the specimen, high-resolution structural features decay exponentially due to radiolysis and bond breakage. Dose weighting applies a frequency-dependent B-factor that optimally down-weights later frames in the movie series. Early frames contribute more to high-resolution terms, while later frames retain value for low-resolution contrast. This statistical weighting maximizes the signal-to-noise ratio of the final averaged micrograph, recovering information that would be discarded by uniform averaging.

Modern frame alignment divides each micrograph into overlapping rectangular patches (typically 5×5 or 7×7 grids) and independently tracks the trajectory of each patch through the movie frames. Cross-correlation or mutual information metrics measure displacement vectors between corresponding patches in successive frames. These measurements are fitted to a smooth, time-varying polynomial model that constrains physically plausible motion. The result is a dense deformation field that corrects anisotropic, swirling movements—particularly important for correcting ice doming near the center of holes in holey carbon grids.

Frame alignment quality is assessed through multiple quantitative metrics. The drift trajectory plot visualizes cumulative X/Y displacement over time, with well-aligned movies showing sub-Ångström residual motion. Per-frame cross-correlation coefficients should plateau after alignment, indicating successful registration. The power spectrum of the aligned micrograph should exhibit Thon rings extending to high resolution without directional anisotropy. Poor alignment manifests as resolution anisotropy in subsequent 3D reconstructions, with preferred orientation artifacts exacerbated by uncorrected beam-induced motion.

A computational compensation applied during frame alignment that optimally down-weights later frames in the movie stack. This accounts for the cumulative loss of high-resolution information caused by radiation damage.

• Applies a frequency-dependent weighting scheme based on critical exposure curves
• Preserves low-resolution contrast from later frames while suppressing noise at high resolution
• Essential for extracting maximum information from direct electron detector movies

A per-particle, beam-induced motion correction algorithm implemented in RELION. It uses a Bayesian framework to model and reverse the trajectories of radiation damage and specimen movement for each individual particle.

• Trains a Gaussian process to model per-particle motion paths
• Outperforms global motion correction for flexible or beam-sensitive complexes
• Iteratively refines particle positions against a running 3D reconstruction

A digital camera technology that directly detects electrons with high quantum efficiency and fast readout speeds. Cameras like the Gatan K3 and Falcon 4 enable dose fractionation by recording movies instead of single exposures.

• Counting mode eliminates Landau noise for improved detective quantum efficiency
• Super-resolution mode captures sub-pixel information
• Frame rates of hundreds of frames per second allow precise motion tracking

A resolution estimation method that splits the particle dataset into two independent half-sets for separate 3D reconstruction. The Fourier Shell Correlation between these maps provides an unbiased resolution estimate.

• Prevents overfitting and noise correlation that inflate resolution claims
• The 0.143 cutoff criterion defines the reported resolution
• Phase randomization of high-frequency shells tests for pure noise fitting

A computational method that calculates a resolution value for each voxel in a cryo-EM density map. It identifies regions of structural flexibility or disorder that are not captured by a single global resolution number.

• Uses a sliding window approach to compute local FSC
• Visualized as a color-coded heat map on the 3D density
• Critical for validating flexible domains, loops, and ligand binding sites