FSCC Drop off at 12A followed by gradual decay

Hey everyone,

I am new the cryoEM/SPARC and would appreciate any help I can get. I am working on a protein that is ~125kDa as a monomer, and 250kDa as a dimer. The dimer is an elongated structure (have crystal structure), while the monomer is globular (no structural information). I am looking at discerning both structures which should be present in the data collected, although the dimer will likely contribute to the majority of the particles it can be induced not only from ligands but also concentration.

I have collected a dataset of 1.3million particles and have done the following general processing.
Started with blob picker, 2D classification, (followed by generous selection) and ab-initio to get ~ 3 classes (dimer, monomer, and some “Junk” reconstruction). I used these inputs to create templates and particle pick. From the new 2D classification (300 templates - 3 rounds to clean up classes) and curation I have approximately 800K particles left over. I have both selected generously and stringently (with high resolution and low ECA cutoffs ~350K particles) followed by ab initio for two classes. These two reconstructions look as expected, there is a high agreement with the dimer crystal structure and the volume of the monomer is consistent what what I would expect.

Regardless of how I treat the data upstream of refinement, the refinements essentially all turn out the same. This would suggest to me that I am doing something wrong or there is a pathological problem with the data (orientations). The refinements have high CC until 12A, where there is a significant drop to ~0.6CC followed by a slow decay to approximately 4A. I assume that there may be some high resolution data present but the medium resolution data is missing or is washed out (poor alignments or?). After refinement the maps look awful, with no high resolution structural data and protrusions around the volume.

I am unsure if there is an orientation problem (probably likely), or if the refinement is having trouble with this dataset and it is potentially fixable. From what I have gathered from the forum, small particles reconstructions can suffer from over-refinement which may also be occurring here. Any insight would be appreciated! I am happy to supply anything that can help.

Hi @mjmcleod64,

I think a poor orientation distribution could be contributing to this. If you look at the slice plots of the refinement job, you can see that the leftmost plot shows high-ish resolution detail, whereas the other two plots show just lower resolution streaks of density, which is a clear sign of anisotropy.

If you have a large particle stack, and you don’t actually have missing views (just overrepresented views), you might be able to improve this by discarding particles from the overrepresented view. The simplest way to do this is to just use select 2D classes on the output of your 2D classification and exclude a bunch of the dominant / overrepresented view. There is also a more automated way to do this, using the Rebalance 2D classes job, which will take the outputs from a 2D classes (or select 2D job) and cluster the classes together based on similarity, then remove particles from clusters that are overrepresented. This will result in a smaller stack, but with a somewhat more uniform orientation distribution.

It would be best to do this with all the particles from 2D classification, then take the new particle stack and move to ab-initio and see if you get an improved orientation distribution, then move again to refinement. Check out the page link for the rebalance 2D classes for a description of the job and parameters.


I totally agree you have a degree of directional anisotropy, but as @mmclean stated you may be able to overcome it by trimming down your large 800k particle stack.

From your screenshot of the blue density (is that the 3.99A map?), it looks boxy and like you may need to change the viewing setting in chimera. Be sure to change the “step” to 1 (top right drop-down of volume viewer tab, or command “volume all step 1”). Sorry if you already know this. It could alternatively be you are using binned particles for the blue density. It looks either heavily binned and close to nyquist or you should change the step to view it properly.