Using 3D template from imported 3D volume for heterogeneous or NU refinement

3D Classification
1

Hi guys，
currently, I am processing one dataset that has heterogeneous particles. I am wondering if I can use good 3D template generated from pdb structure by chimera for heterogeneous refinement or NU-refinement.

2

Would strongly recommend against this due to the risk of template bias.

2 Likes
3

Hi zcp, Agreed with olibclarke on this, is this a dataset where you are struggling to even get something that resembles the protein at low-res from ab-initio?

4

Thanks. Is it possible to integrate function like relion in the future ?

5

yes. Ab-initio does not give good template.

6

By “function like relion”, what do you mean? This is possible in both software, but strongly discouraged in both to my knowledge.

7

And how do your 2D classes look? Clear secondary features, or blobs? How did you pick your particles? Can you clearly see them on the micrographs?

8

it shows very clear secondary feature, but it has some preferred orientation issue. when i tried ab-initio, it failed to generate good template. in relion, you can import one template for 3D classification. I thought this may be helpful for addressing issue like this one.

9

You can import templates - just be aware that there is a risk of template bias, and use caution. If you do this, I would strongly recommend aggressively lowpass filtering the template prior to using for refinement (say 40 Å or so).

Usually, there are settings in ab initio which can be adjusted to mitigate issues like what you are describing. If you see clear features in 2D, a good ab initio model will usually be obtainable. What did you try?

10

I just used the defaut setting for ab-initio. Could you let me know what setting i can adjust?

1 Like
11

what size is your particle? If it is small, and has clear secondary structure features, then I would recommend using a higher resolution range (say 9-7Å).

Switching off both “recenter in real space” and “enforce non-negativity” can help in situations like you describe.

Also decreasing the Fourier radius step by 5-10x will often give better, more stable results with less false minima/artefacts, at the expense of longer runtimes.

If you search the forums, you will find more detailed discussions of each of these points (except perhaps the last one)

12

really appreciate it. size is 400 kd. I would try like you recommend.

1 Like