I have collected a dataset from my protein. I can see different angles of the protein in 2D classes. The data has been collected from a GO grid. I have a good number of particles after 2D classification and I have tried to clean it as much as I could by running various 2D classes. When it comes to Ab initio and building the 3D model, it gives me a weird model, it is like different part of the protein are not connected, it shows the protein in different layers. Does anyone have any suggestion how to change the setting to improve the 3D model? I usually use the default setting and only change the max res and initial res to 3 and 10, because my protein is a small membrane protein, 150 kD. I have resolved the structure of this protein before, from a different dateset and I am using the same setting as the previous one.
max res of 3 is very high - can work sometimes but is usually unnecessary, and can lead to the solution falling apart into noise. I would try an initial res of 9 and final res of 7. Can you post a screenshot? It is hard to tell what the issue is exactly from the description.
If you have a lot of particles and are going to 9,7,3A or so (I would not usually go under 5-6A and default 12 is ok) - then having several classes for ab initio is not a bad idea. With class similarity set to something reasonable you can already use ab initio to start getting out junk of different conformations.