Weird result of NU refinement

3D Refinement
1

Dear community,
I am working on a dataset of a 120kDa complex. Homogenuous refinement gives me resolution of around 7A while NU refinement improves the resolution to 5A. However, the density I obtain from NU refinement does not show any details or features, and the density I obtain from homogenuous refinement looks normal. Please see the screenshots of the NU-refinement densities and the associated FSC curve.

image
image766Ă—854 136 KB

image
image691Ă—785 72.1 KB

image
image784Ă—426 54.8 KB

Can anyone please explain to me what causes this issue and whether there is anyway to avoid /overcome this to get better results?
Thank you so much!
—Da

2

What did the density look like in the corresponding homogeneous refinement? Could you see clear secondary structure? How did the 2d classes look?

2 Likes
3

A representative micrograph too (raw and denoised, if possible).

2 Likes
4

The 2D looks pretty good.
Here’s a screenshot of the density from homogenuous refinement at a resolution of 7-ish angstrom. It looks normal to me.

image
image834Ă—1095 134 KB

image
image845Ă—1052 112 KB

5

This does not look normal. It also looks like junk unfortunately

6

when low-filtered to 10 A, I can fit the atomic model into the density and the density shows clear feature.
Do you think I should further clean up the particles to remove trash? d

7

here’s a sample micrograph.

image
image945Ă—671 145 KB

8

It really doesn’t have clear features I’m afraid. How do the 2Ds look?

9

(post deleted by author)

10

(post deleted by author)

11

The template I used is a published pdb. Somehow Topaz does not work pretty good in this dataset.

12

I would strongly recommend avoiding using a pdb as a template. What do 2Ds look like starting from blob picking?

13

I didn’t use blob picking because the person who processed the dataset before me told me blob picking did not yield good result.
I just launched a blob picking job to see how the 2D looks like.
Does picking with pdb as template yield bias in picking (and not picking “good” particles)?

14

Does picking with pdb as template yield bias in picking

It can, although having said that if you filtered them appropriately & used appropriate cutoffsz during particle curation it shouldn’t be a major issue. What resolution did you filter the templates to?

If you filtered them to 20Ă…, then these 2Ds are encouraging, but I would say something has gone wrong during either initial model generation and/or refinement.

I would also suggest trying Micrograph Denoiser prior to blob picking.

15

I filtered to exactly 20A. The reference I used during refinement is the previous EM structure.
I will try micrograph denoiser and then blob picking to see whether the result improves.

16

Maybe this is your issue - did you try starting from an ab initio model generated from your particles?

17

No I didn’t! I will generate ab initio model from the 2D to see whether using it as reference can make refinement better. Thank you so much! I will keep the result once it finishes.

1 Like
18

Launching refinement job now. The ab initio reference looks good to me. I can fit the atomic model into the 12 A reference with no difficulty.

image
image684Ă—685 40.8 KB

19

Ok good luck! I would also check your gain reference - there is a distinct horizontal line in the micrograph you attached which shouldn’t be there if gain correction is optimal

20

Yes I also realized it a short while ago. “Flip y” would make it disappear, which means maybe I need to start over

1 Like