Unreliable all GSFSC curves after homo/hetero refine, why?

3D Refinement
1

hello, everyone!

i have a quection about weird GSFSC curve.

my target is a membrane protein, there appears a strange GSFSC curve(all curves) when i did the NU refine or homo refine job with the particles from hetero refine or ab_initio job, just shows below.

WechatIMG19314
WechatIMG193141046Ă—590 129 KB

so, i want to ask why and how to deal with it

thanks a lot~

2

Have you checked for duplicate particles? The fact that the curves are not returning to zero often signals the presence of duplicates.

Also, the sharp drop in the corrected curve often results from either still having a lot of junk in the particle set, or using a refinement mask which is not soft enough.

Cheers
Oli

3 Likes
3

[quote=“olibclarke, post:2, topic:13176”]
Also, the sharp drop in the corrected curve often results from either still having a lot of junk in the particle set, or using a refinement mask which is not soft enough.
[/quote
yes, you are true! i have removed duplicate particles. and also done some refine job. the curves did return to zero. here, i also have some questions during these processes

  1. as for removing duplicate particles, there some parameters need to change, how can i determine the right parameters?

    image
    image1488Ă—598 99.6 KB

    such as score field, shift key?

  2. when i finished these steps, the resolution is decreased from 3.7 to 4.3 angstrom, how can i improve the resolution(i have tried running ab_initio job to get more pure particles, but it seems did not work)

4

Distance for duplicate removal depends on the size of your particle, and how densely packed they are in the micrographs (also a little on shape, although this should be considered a subset of particle density). The sign that there are no duplicates (at least shared between half-sets, although symmetry expansion is currently outside of the scope of this troubleshooting) is that the FSC curve falls to zero and stays there. I usually leave the score field at the default setting.

Your resolution decreased because you (probably?) have a healthy FSC curve which falls to zero, so aren’t artificially boosting correlations between the half-sets by duplicating particles between them. :wink: You can try creating some projections of the 3D, re-picking, classifying and combining (with remove duplicates to make sure you aren’t double picking!) but it depends on the quality of your micrographs, biochemical prep, etc. :slight_smile:

1 Like
5

Hi olibclarke,

Thanks. I have a question about how to figure out which possibility is true. Also, why does using an unsoft refinement mask lead to this situation? What’s the mechanism behind this? And how to make a soft mask for the refinements (Nu-refine or homo refine job)?

Looking forward to your response,
Jianming

6

Hi Jianming,

User DanielAsarnow has a nice explanation for why you need a soft mask for refinement in this post.

You can also find more information in the mask creation tutorial page.

Essentially, masking a volume converts all values outside the mask to zero. Trying to use Fourier components to approximate this steep drop-off in voxel intensities from inside to outside the mask results in artifacts, especially if the mask cuts through protein.

Best,
cbeck

7

Thank you very much, I will check.