Advice on Ab-Initio related experience

3D Refinement
1

Hello everyone, I would like to ask if the particles selected by 2D classification can be directly Ab-Initio? Thank you all for your help.

J229_2d_classes_for_iteration_60
J229_2d_classes_for_iteration_602409Ă—1208 440 KB

J230_2d_classes_for_iteration_61
J230_2d_classes_for_iteration_612409Ă—1208 440 KB

J231_2d_classes_for_iteration_61
J231_2d_classes_for_iteration_612409Ă—1208 441 KB

J227_2d_classes_for_iteration_60
J227_2d_classes_for_iteration_602409Ă—1208 423 KB

J228_2d_classes_for_iteration_60
J228_2d_classes_for_iteration_602409Ă—1208 438 KB

2

Yes. :slight_smile: Use a reasonable number of classes (8-24).

1 Like
3

Ok, because there are some vague 2D drawings, so I wonder if it will affect? Thank you.

4

The variable quality of the 2D is why I suggested so many classes. :slight_smile:

5

Thanks for your advice. I’ll try it.

6

It doesn’t look very big, I’d also try increasing minibatch size and decreasing starting and target resolution (maybe 20 start, target 8?)

7

looks like you have a very small amount of quality data, though the protein sample is quite nice. I would try to collect more with a new grid. You could also try better 2D with batchsize 1000 and 40 o-EM iterations. Try 2D select only the GREAT ones (first couple classs each 2D) and then inspect picks with that input to find out which micrographs/regions of micrographs give the great particles. Is there consistently worst ice everywhere else? Or are good particles only in micrographs with limited motion? Or are good particles only at the edges of the micrograph and never the middle (or inverse)?

I would also take all the GREAT classes of each “view” and train individual Topaz pickers then repick all micrographs with each Topaz model and combine/remove duplicates.

8

I’ll try. Thanks for the advice.

9

Hello,

I guess you might already have some results from ab-initio? Do you have any nice-looking envelopes? You might consider treating this with the special workflow for flexible+small proteins - look for it in this forum, there are a few topics on it.

10

Thank you for your advice. At present, the resolution of the structure I got is about 6 anglais. I wonder if there is any way to improve it.

11

Linqian,

If you are getting what I think you are getting, I would not look at resolution at this point. Does your map show internal features? At 6 angstroms you should be able to distinguish helices at least.

1 Like
12

Agreed. One of our postdocs excitedly reported a 5.6(something) Å map last year from a very flexible complex… yeah, it’s about 12 Å, at best. Probably more like 15-20Å. When I say “blob territory”, I’m not joking. Although their 2D wasn’t as nice as OPs.

1 Like
13

Thank you for your mention, the program gives a resolution of about 3 angstroms, but the actual resolution should be 6-8 angstroms