Random-Conical Tilt (RCT) Reconstruction workflow¶

Purpose: Basic workflow for processing an RCT data set.

see also RCT run protocol from a user

General Workflow:¶

1. Pick particles: Pick particle on the tilted and untilted micrographs. This can be done in separate jobs (e.g. with multiple templates), or in one job. If your particles are small, one strategy might be to pick more particles than are clearly evident. The reason for this is that particle pickes will get "filtered" (or thrown out) during the tilt-pair alignment if the tilt pairs don't match up, as well as during classification.
2. Align tilted and untilted particles: The first step might be to try the Auto Align Tilt Pairs. If you have good overlap of your imaged area in your tilted and untilted micrographs, as well as picked particles on both tilt-pairs, then the auto-tilt alignment should work without problems. If, for example, you have little overlap or few particle picks, then the auto-tilt alignment might fail, or, worse, align the wrong tilted particle to its untilted counterpart. it is therefore very important to check the result of this step in the imageviewer. If the auto-tilt aligner fails, then the particles must be aligned manually using the manual Align and Edit Tilt Pairs function before proceeding to the next step.
3. Create tilted and untilted stacks: Create (1) an untilted stack from the untilted particle picks and (2) a tilted stack from the tilted particle picks, referring to the picked set made using the alignment step above. The two stacks should have roughly the same number of particles. Note that they rarely have an identical number of particles, because some particles get rejected if they lie too close to the edge. This will not make a difference in the later stages of processing. The important thing is to have a tilted and untilted particle stack from the tilt-pair alignment step above. Once the tilted stack exists, you can leave it alone. You can manipulate the untilted particles as you wish (e.g. align, classify, remove particles, etc.). Since each untilted particle has a database record that corresponds to its tilt-pair, as long as the information stays within Appion, you can proceed without the need for any extra bookkeeping.
4. Align (and classify) untilted particles: The untilted particle stack must be aligned. You can use any Particle Alignment available (however, note that, CL2D works with RCT in Xmipp 3.0, but not in Xmipp 2.4. In Xmipp 2.4 CL2D does not save alignment information into a docfile, and therefore CL2D cannot be directly used for RCT reconstruction). If the alignment procedure also contains a classification / clustering within it, then the resulting class averages can be directly used for RCT reconstruction. Otherwise, you must classify / cluster the aligned stack, e.g. using Correspondence analysis, followed by hierarchical ascendance classification, or, e.g., Xmipp Kerden Self-Organizing Map.
5. Reconstruct each class: Each class of particles produced by the alignment and classification step can be used to create an RCT volume. If you display the resulting classes in the viewer, you can click on any number of them, then click on 'create RCT volume', which will take you to the RCT Volume launch page. Once there, fill in the required information to create a map. If you click on more than one class, then the resulting volume will be a combination of the particles in those classes. However, if the two are NOT aligned in-plane with respect to each other, then they should not be combined, because the Euler angles will be averaged out.
6. Assess the accuracy of the classification: Several ways to do this. First, one should make sure that the aligned particles actually belong to the class from which the volume was reconstructed. Most, or at least the majority, of the particles should actually look like the class average. Otherwise, there is too much heterogeneity in the class, and the classification and RCT reconstruction should be redone. In general, it is a good idea to create as many class averages as possible to accurately capture all the different views in the data without sacrificing signal-to-noise ratio. This is not a straightforward task, but can be approximated fairly well. As a general rule of thumb, ~100 particles per class is sufficient to provide a class average with good signal. ~500 particles per class should provide enough particles for a 40Å RCT map.
7. Assess veracity of the RCT map: RCT reconstruction is a very powerful strategy for reconstructing electron density maps, particularly of heterogeneous systems, but also has its limitations. First the resulting volumes contain a missing cone of information (Radermacher, M., T. Wagenknecht, et al. (1986). "A new 3-D reconstruction scheme applied to the 50S ribosomal subunit of E.coli." Journal of Microscopy 141: RP1-RP2 and Radermacher, M., T. Wagenknecht, et al. (1987). "Three-dimensional reconstruction from a single-exposure, random conical tilt series applied to the 50S ribosomal subunit of Escherichia coli." J Microsc 146(Pt 2): 113-136) and second, it may be severely flattened (Cheng, Y. F., E. Wolf, et al. (2006). "Single particle reconstructions of the transferrin-transferrin receptor complex obtained with different specimen preparation techniques." Journal of Molecular Biology 355(5): 1048-1065). To minimize the effects of the missing cone and particle flattening, you can average different views of the same particle. The latter must be emphasized, because if your particles are either conformationally or compositionally heterogeneous, then you additionally have to determine whether or not distinct class averages arise from particle heterogeneity OR different views of an otherwise identical object. If they represent different views of an otherwise identical object, then, if you align and average in 3D the different resulting RCT reconstructions of distinct views, you should in theory come up with an improved result. If they represent heterogeneous specimens, then aligning the resulting volumes in 3D will only make things worse, because any information that can be potentially gained from distinguishing heterogeneity will be lost to averaging. Discriminating between these two possibilities is not trivial, but can be done with careful analysis of your data. In the case of a single preferred orientation, then the resulting map will be flattened, AND it will contain a missing cone of information. Although this cannot be avoided, the latter scenario may or may not affect the interpretation of the map, depending on what is desired.

< Step by Step Guide to Appion | Processing Cluster Login >