I am a newer Cryosparc user, attempting to refine a helical structure. I am wondering the best way to start the pipeline and if there is something I am doing wrong with the ab initio job.
I imported particles and then run multiple Class 2D jobs and selections, ending up with approximately 220k particles of reasonable quality. I then hand these particles to the ab initio job with 5 classes, hoping to further sort out junk particles. Unfortunately it seems that the output quality deteriorates as the job progresses. In the first iterations the filament appears to align as expected in the XY view, but this is lost as the job proceeds, resulting in nonsense outputs. Can anyone suggest a reason for this or an alternative workflow to further clean the dataset?
We have had some success with doing an initial homogeneous reconstruction and then using this as an initial model for asymmetric helical refinement and subsequent searches of helical parameters, but I feel there are some gains to be made with a cleaner dataset.
Ab-initio reconstruction for helical particles depends significantly on the dataset. For some datasets with helical symmetry that have enough low-resolution features (one example is EMPIAR-10495), ab-initio works excellently because it can use those low-resolution features (that are usually present in globular proteins) to guide search in the early iterations. However, for helical particles with small ASUs (e.g. tobacco mosaic virus), ab-initio exhibits a failure mode that often causes the density to “collapse” from a cylindrical shape to a flattened shape.
We have observed that with some datasets, adjustments to ab-initio’s “Enforce non-negativity” and “Center structures in real space” parameters can improve results. It is worth checking out the suggestions of this thread for more details. Setting the initial and maximum resolution parameters to lower numbers (smaller wavelengths) may also help. But, in our experience, with helical particles the general recommended workflow in these cases is to directly begin with helical refinement, and to generate an initial density via a cylindrical one (this can be done in the “Initial model” section of the helical refinement parameters).
Thanks for the quick response. It seems that if I do the non-negativity and center structure tweaks, as well as drastically decrease the target resolutions I can get closer to a reasonable outcome, but still not particularly useful.
What I’ve tried instead is to do a homogeneous reconstruction directly of my 2D classes, which gives me a cylinder with a lumenal density, which I think might be a slightly more accurate template than a ‘pure’ cylinder.
After helical refinement, I see a strange behaviour of the FSC, where the tight mask diverges from the rest early in the process and therefore reports incorrect resolutions. After reading the forums, it seems this may be caused by incorrect behaviour of the dynamic mask, so this is switched off by setting the start resolution to 1A. I also restrict the alignment resolution to 10A. The gross morphology of the structure is similar to what we expect.
Is a reasonable next step here to search for symmetry around the central Z slices of the structure and then move towards local refinement to avoid these FSC/masking issues?