Anisotropy from single class signal dominance

Hello all,

I’ve refined my map to a decent resolution but there’s obvious anisotropy, especially at the exterior. I used NU-refinement with local and global CTF refinement enabled.

It doesn’t seem that the anisotropy is caused by preferred orientation because the class causing it has about the same number of particles as the others. It appears to me that it is caused by dominance of the information coming from the class, secondary structure is much clearer in this class than the others and the anisotropy is in the direction of this particular class.

I’ve been told to exclude the class altogether but I lose about 2A resolution and there is no density in the regions I’m most interested in.
I’m thinking that I should apply a low pass filter to exclude some of the high resolution information coming from the class but I’m not quite sure how to go about this without sacrificing too much global resolution?

I’ll appreciate any ideas.

Thank you.

First of all, you should consider what you want to get out of your dataset - if you have better signal from a certain direction no matter the reason, the optimal outcome would be an anisotropic map. If you want to get a more isotropic map at lower resolution and you think that the refinement runs well for intermediate iterations, it should be a simple matter to limit the alignment to a lower resolution or just low-passing the map after the fact. If it seems that the refinement is overfitting because of the anisotropy, and you want to try a variant of what you’ve already tried, I would try another 2D Classification to split up the good 2D class and bring along some of the particles instead of discarding them completely as you describe.
I should also say, however, that even though it doesn’t seem due to preferred orientation based on the population of your 2D classes, the maximum likelihood class assignment will put more junk into your worse classes, so if the worse projections do not completely lack low-resolution features and the dataset isn’t very, very clean, it may seem like you have equally distributed views, even if you don’t. To get an idea of whether collecting new data is worth your time, DeepEMhancer can, allegedly, often predict what your map could look like given data to extend the high resolution in bad directions, so maybe that’s worth a look.

Sincerely, Martin


Hi Martin,

Thanks for the insight. I’m relatively new to SPA, so I appreciate the feedback.


Hi Martin,

So after taking a few steps back I realised that the stretched effect on my map wasn’t as a result of anisotropy. I get it when I do a NU refinement with local and global CTF enabled.

I think it’s sorted but it would be interesting to know why I get the stretched appearance after NU refinement.


In general I would say that a stretched map is exactly anisotropy, but the cause of it may not necessarily be preferred orientation. My best guess would be that homogenous refinement with default parameters might stop refinement at a lower resolution before the anisotropy becomes severe, whereas the extra degrees of freedom of the NU refinement with CTF refinement may take refinement slightly further to a point where the anisotropy becomes obvious. To get a better idea of what is causing the anisotropy, submitting the half-maps to 3DFSC or the angular assignments to cryoEF might give you some more insight into your data.

Sincerely, Martin

I would double check the log during CTF refinement - if resolution is low both local and global CTF refinement can fail and make things worse.

@au583982 the 3dfsc map has a bit of density missing in the middle of what I expected to be a complete sphere. So there is anisotropy.
I think you are spot on regarding the NU refinement with CTF refinement. The map gets much worse with NU refinement even without CTF refinement. I’ll try DeepEMhancer too. Unfortunately, collecting a new data set is not an option so I’ll refine with limited resolution and maybe Fourier crop and compare the resulting maps.
@olibclarke I was aware that CTF refinement could worsen resolution for low resolution maps but mine is sub 3A without visible stretching, that’s why I thought that I could do CTF refinement. I’m not sure if 3A qualifies as low in this case.

Thank you both!

If it is sub 3Å I would agree it’s less likely, though refinement of some params (e.g. Cs) can still be unstable at relatively high resolution. If the issue occurs without CTF refinement then I would tend to agree that limiting res for alignment is the way to go

Also, is this a multidomain protein? If so, consider the possibility that anisotropy may be due to interdomain mobility, rather than preferred orientation, in which case local refinement may be beneficial.

Its a homo-oligomer with C2 symmetry
How would you tell what the source of the anisotropy is and would the strategies to fix it differ based on where it’s coming from?

How big is the monomer? If it is big enough (~>50kDa), I would try performing C2 symmetry expansion after refinement with C2 symmetry, and then local refinement in C1, with a mask around a single monomer. If the anisotropy is due to flexibility at the dimer interface then that may help.

1 Like

So the monomer is around 38kDa but the main building block of the whole thing is the dimer, I haven’t seen any evidence yet that it exists as just the 38kDa monomer.

Thank you so much, I’ll give that a try.
I might also attempt refining on Relion, maybe their algorithm might handle the anisotropy better.

1 Like

Doesn’t really matter whether it exists in solution as a monomer - there can still be flexibility between the two monomers in even an obligate dimer, so I would give it a go