MotionCor2

MotionCor2 performs GPU-accelerated motion correction of dose-fractionated cryo-EM image stacks, compensating whole-frame drift and anisotropic local beam-induced sample motion to improve image quality for high-resolution reconstruction.

Key Features:

• GPU acceleration: MotionCor2 is GPU-accelerated and can execute across multiple GPUs on Linux systems.
• Motion model: Models non-uniform motion as smooth, time-varying 2D polynomial fields resulting from projections of underlying 3D specimen deformation during electron exposure.
• Whole-frame and local correction: Simultaneously compensates uniform whole-frame drift and anisotropic local beam-induced sample motion.
• Patch-based estimation: Partitions each movie frame into spatial patches and iteratively estimates local inter-frame shifts.
• Polynomial trajectory fitting: Fits local inter-frame shifts to continuous polynomial motion trajectories over time.
• Pixel-resolution remapping: Remaps each sub-frame at pixel resolution using the inferred deformation fields.
• Weighted summation: Sums remapped sub-frames with optional dose- or B-factor weightings.
• Bad-pixel handling: Integrates bad-pixel detection and correction.

Scientific Applications:

• Archaeal 20S proteasome: Benchmark micrographs show sharper Thon rings, improved CTF agreement, and higher nominal 3D reconstruction resolution relative to MotionCorr and Unblur.
• TRPV1 ion channel: Benchmark micrographs of TRPV1 demonstrate improved image quality and reconstruction metrics consistent with the proteasome results.
• Cryo-electron tomographic tilt series: Demonstrates robust performance across tilt-series and low-defocus data, with improvements comparable to workflows that include RELION particle polishing.
• Preprocessing for high-resolution reconstruction: Used as a preprocessing step to produce near-atomic-resolution reconstructions from dose-fractionated movies, often with minimal additional gain from particle-level trajectory refinement.

Methodology:

MotionCor2 partitions each movie frame into spatial patches, iteratively estimates local inter-frame shifts, fits these shifts to continuous polynomial motion trajectories over time using a deformation model of smooth, time-varying 2D polynomial fields derived from projections of 3D specimen deformation, remaps each sub-frame at pixel resolution using the inferred deformation fields, and sums frames with optional dose- or B-factor weightings while integrating bad-pixel detection/correction and leveraging multi-GPU execution.