Motion Correction
Last updated
Last updated
During imaging in the microscope, a cryo-EM sample is irradiated by the electron beam for typically between one and ten seconds. During this time, the sample does not remain perfectly still.
Drift of the stage, vibration of the microscope, and deformation of the sample ice all contribute to motion of the particles. These effects are all visible as motion blur in the final recorded image. To allow for correction of this effect, microscope images are captured as multi-frame movies (typically approximately fifty frames over the entire exposure). Because each frame only captures a short amount of time, the motion blur in each frame is significantly lower than if the entire micrograph was collected at once (i.e., if there was only one frame).
Motion correction is the process by which those frames of a raw movie are aligned and averaged to produce a single-frame micrograph. This process significantly improves signal-to-noise ratio over collecting data in a single frame by reducing the cumulative effect of motion blur.
In addition to causing motion, the beam interacts directly with the sample. The electron beam is a powerful source of radiation, capable of damaging the sample. High-resolution features are especially sensitive to radiation damage. Since the late frames have received the greatest radiation dose, they also tend to have the lowest quality information at high frequencies.
A technique commonly called dose weighting (Grant and Grigorieff, 2015) accounts for the varying information content in each frame by attenuating the high-frequency signal from later frames of movies. All motion-correction jobs in CryoSPARC apply dose weighting. See the relevant section of this page for more information on the specific forms of dose-weighting applied by each job.
CryoSPARC provides multiple motion correction methods and workflows. In almost all projects, Patch Motion Correction should be used in the initial processing steps. It is also used internally by CryoSPARC Live when performing real-time processing. If the final 3D reconstruction is of high quality, Reference Based Motion Correction may provide additional resolution improvement.
Motion correction is the process of algorithmically correcting for motion of the electron microscope stage and the sample ice itself to recover image quality lost by motion blurring.
Cryo-EM data is collected in the form of movies, which are each a series of individual frames. Since a frame is usually between 0.1 and 0.2 seconds, the detector does not accumulate enough electron dose for clear identification of the target. However, the brief length of time significantly reduces the amount of in-frame motion blur.
Since the same physical objects create the image in each frame, we can find a shift to apply to each frame (or sub-region of a frame) that results in the greatest agreement among all frames. The ultimate goal of motion correction is to reduce the total blurring in the final particle images that are extracted — what differs between the different methods and implementations is the type of input data they require and the types of motion they are capable of capturing and correcting.
There are two main forces which cause motion during data collection: stage drift and beam-induced ice deformation.
The cause of the former is self explanatory: mechanical effects cause drift of the entire stage and grid during movie collection. This motion is observed as a shift of the entire image frame, and is the easiest type of motion to model. Each frame can simply be translated to create the best match between the previous and the next. This is called Rigid or Full Frame Motion Correction, because it produces a movement trajectory of the entire frame as a rigid object. Note that rotation of the grid is generally negligible and is not modeled.
When the sample is irradiated by the electron beam, the thin layer of ice suspended in the grid hole buckles. This three-dimensional movement appears in the movie as anisotropic (i.e., different at different spatial positions) movement of the ice itself. The exact mechanism behind this effect is not fully understood, but it is suspected that the electron beam allows for relaxation of physical stress built up during sample vitrification (Thorne 2020). Modeling this type of motion is more challenging, since in theory small image regions in each frame might move in a different directions. CryoSPARC models anisotropic motion primarily using the Patch Motion Correction job, and the theory and method underlying that job is described in the job page.
As discussed above, the goal of motion correction is to align each frame such that the particles are in the same position throughout the movie. This way, when they are averaged, the high-resolution information is not lost due to motion blur. For example, consider the movement of particles move throughout this movie:
If this movement is not corrected, the particle images would be unacceptably blurry, destroying high-resolution information:
The main source of motion for a given particle is typically concerted motion of the entire stage, or frame. It is relatively straightforward to correct for this type of motion by minimizing the difference between one frame and its neighbors. In doing this, each frame is brought into register with the others. This corrects the rigid movement of the entire frame, while leaving the anisotropic movement unmodelled:
While this unmodelled movement still results in blurry particles, the results of averaging these aligned frames is already much clearer than if the movie had been collected as a single-frame micrograph:
As this is a relatively simple form of motioncorrection, it was one of the earliest forms introduced. Some examples of early implementations of Full Frame Motion Correction are MotionCor (Li et al. 2013) and Unblur (Grant and Grigorieff 2015).
Correcting the anisotropic motion of individual particles is more challenging than the full-frame motion for two main reasons. First, and most importantly, single particle images each contain far less signal than an entire micrograph. Second, correcting for anisotropic motion requires knowing the position of particles to begin with. Once good particle location information is available, there are two types of jobs to correct the movement of individual particles in CryoSPARC.
The first is Local Motion Correction. In this job, a small patch around each particle is compared in each frame and aligned so as to reduce the total motion, much like the process in Full Frame Motion Correction. The second type is Reference Based Motion Correction, which compares a projection of a high-quality 3D Volume to the particle location in each frame to find that particle’s exact location.
In general, with a high quality reference volume, we expect Reference Based Motion Correction to perform better than Local Motion Correction. Reference Based Motion Correction also estimates empirical dose weights (see the Dose weighting section) and is based on Bayesian Polishing (Zivanov et al. 2019). However, the requirement of a high-quality volume is significant. Local Motion Correction does not require a reference, and so can be run early on in the processing pipeline if significant anisotropic motion is observed. Local Motion Correction is based on alignparts_bfgs (Rubinstein et al. 2015).
Patch Motion Correction is a fourth motion correction job available in CryoSPARC. It provides a means of correcting anisotropic motion without knowing particle positions. To do this, it models the movement of large patches of the micrograph, then describes that movement using a function called a spline. Importantly, this spline can be evaluated at any pixel location to find the trajectory of that pixel during the recording of the movie.
When a particle image must be extracted from a position in a Patch Motion corrected micrograph, the spline function is evaluated at each pixel position to create an aligned average for that pixel. In this way, anisotropic motion is corrected without knowing the particle locations before hand. Therefore, this job typically gives better results than Full Frame Motion Correction, and is the job we recommend for motion correcting any new dataset in CryoSPARC. More information on this algorithm is available in the Patch Motion Correction job page.
During image collection, samples are bombarded with an intense electron beam. This beam damages the fragile bonds in the macromolecule, with the amount of damage increasing as electron dose accumulates. In 3D reconstructions, this is visible as a degradation of the high-resolution signal in later frames of the movie — the high resolution information has been burned away by the electron beam.
To alleviate the worst effects of this radiation damage, the process of dose weighting involves down-weighting the contribution of late frames to the micrographs. In this way, it is possible to both
retain the low-frequency signal from late frames, which is less damaged by radiation and useful for particle picking
discard the high-frequency noise from late frames (since there is almost no useful information at these frequencies).
We can plot the dose weights as a series of bar graphs in which the first frame is the topmost bar and the last frame is the lowest bar, and the weight of a frame at a particular resolution is given by the length of the bar.
When plotted this way, the trend aimed for by dose weighting becomes apparent. At low resolution (left panels), all frames are more-or-less equally reliable since the effect of the electron beam is much less noticeable at this resolution. Therefore all frames are treated equally in the final micrograph — they all have a weight of 1.0.
However, radiation rapidly damages information at the highest frequencies (right panels). We therefore want to use only information from the early frames at the highest resolution, so early frames have a weight greater than 1.0 and late frames have a weight less than 1.0.
Visualizing dose weights in this way can become unwieldy when considering all frequencies in a movie. We therefore typically present them as a heatmap instead, where the columns correspond to a resolution, the rows correspond to frames, and the color denotes the weight associated with that frame at that resolution.
The heatmap above shows an example of the default dose weights applied to movies during motion correction. These default dose weights are calculated in the same way as described by Grant and Grigorieff. These dose weights are based on a model of exponential decay of the signal-to-noise ratio and are applied without any knowledge of the underlying sample or movies.
Default dose weights work well in most cases, but if more information about the system is available a better estimate of radiation damage can be derived. More specifically, if the position of particles in each frame, high-quality pose estimates, and high-resolution reference volumes are available for each particle, it is possible to calculate the correlation between the reference volume and the particle image. From these correlations we can deduce the appropriate dose weights for each frame and resolution.
These calculated dose weights are called empirical dose weights and can be calculated by comparing a 3D reference with the particles in the movie. This process is performed, for example, by Bayesian Polishing in RELION (Scheres 2014; Zivanov et al. 2019) and Reference Based Motion Correction in CryoSPARC. More information about the procedure is available in that job page.
Thorne, R. Hypothesis for a mechanism of beam-induced motion in cryo-electron microscopy. IUCrJ vol. 7 416–421 (2020).
Li, X. et al. Electron counting and beam-induced motion correction enable near-atomic-resolution single-particle cryo-EM. Nature Methods 10, 584–590 (2013).
Grant, T. & Grigorieff, N. Measuring the optimal exposure for single particle cryo-EM using a 2.6 Å reconstruction of rotavirus VP6. eLife 4, e06980 (2015).
Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nature Methods 14, 331–332 (2017).
Nakane, T. et al. Single-particle cryo-EM at atomic resolution. Nature 587, 152–156 (2020).
Scheres, S. H. Beam-induced motion correction for sub-megadalton cryo-EM particles. eLife 3, e03665 (2014).
Zivanov, J., Nakane, T. & Scheres, S. H. W. A Bayesian approach to beam-induced motion correction in cryo-EM single-particle analysis. IUCrJ vol. 6 5–17 (2019).
Rubinstein JL, Brubaker MA. Alignment of cryo-EM movies of individual particles by optimization of image translations. J Struct Biol (2015).
