MotionCor2

Estimates and corrects the overall translation and rotation of the entire micrograph across all movie frames. This step removes the dominant, large-scale drift caused by stage instability and initial beam-induced movement. The algorithm uses a cross-correlation approach to align each frame to a running average, iteratively refining the alignment until convergence. This single, rigid-body transformation is applied uniformly to the entire field of view before local refinement.

Divides the micrograph into a grid of overlapping patches and independently tracks the trajectory of each patch across frames. This models non-uniform, anisotropic motion caused by beam-induced doming, bubbling, or warping of the vitreous ice. The local shifts are fitted to a time-varying polynomial model to produce a smooth, per-pixel motion trajectory. This step is critical for recovering high-resolution information at the edges and corners of the image.

Applies an optimal exposure-dependent weighting to each movie frame before summation. Later frames receive lower weights because radiation damage progressively destroys high-resolution structural features. The weighting scheme is derived from the cumulative electron dose and the measured critical exposure curve of the specimen. This maximizes the signal-to-noise ratio in the final, motion-corrected micrograph, preserving low-resolution contrast while optimizing high-resolution detail.

Leverages NVIDIA CUDA to parallelize the computationally intensive cross-correlation and polynomial fitting operations. By offloading these tasks to the GPU, MotionCor2 achieves real-time processing speeds, correcting a full movie stack in seconds rather than minutes. This acceleration is essential for high-throughput cryo-EM pipelines where thousands of movies must be processed during data collection. The implementation uses FFT-accelerated alignment on the GPU for maximum throughput.

Generates several critical outputs for the single-particle analysis pipeline:

• Motion-corrected micrograph: The dose-weighted sum of aligned frames.
• Per-frame shift traces: X/Y translation plots for global and local motion diagnostics.
• Local motion map: A vector field showing the magnitude and direction of local movement.
• Defect and outlier pixel maps: Identifies bad pixels or detector artifacts for masking. These outputs are directly compatible with RELION, cryoSPARC, and other major processing suites.

Corrects for geometric distortions introduced by the electron optics and detector, including elliptical distortion and magnification anisotropy. These systematic errors cause the image to be stretched or compressed differently along orthogonal axes. MotionCor2 applies a calibrated affine transformation to restore the true geometry of the specimen, which is essential for accurate particle alignment and high-resolution 3D reconstruction. The calibration parameters are typically derived from a known reference standard like a gold cross-grating.