Skip to Content

Refinement strategies

8 replies [Last post]
mtherkel
Offline
Joined: 19 Apr 2015

Hi all,

I generated an initial model for an icosahedral virus following the scheme outlined in the 2016 Frealign paper. It's working great so far, but I have some general questions on strategy. After 16 cycles of MODE 3, and with a 30 A resolution limit, my FSC now extends beyond 10 A. Do you recommend switching to MODE 1, or should I continue to increase the resolution limits while in MODE 3?

I'm also unclear of how to approach classification. I currently have 95% occ in one of my three classes. Do you recommend seeding my next cycles with this reconstruction as I increase resolution limits? How do you decide when to seed?

Thanks for your help,
Matt

niko
Offline
Joined: 5 May 2009
You should be able to switch

You should be able to switch to Mode 1 after just two or three Mode 3 cycles, especially if the resolution does not improve further. Having 95% of the particles in one class suggests that your dataset is pretty homogeneous, as one would expect for an icosahedral particle (not much room for variability, unless there are some conformational changes). In fact, the 5% of particles that went into the other classes may just be due to noise and inaccurate classification, not true structural differences. You can test if the 95% of particles represent just one structure by removing the 5% of particles that were assigned to the other classes, then seed again three classes using the 95% of remaining particles. They will hopefully either generate again one dominating class, or two or three equivalent classes. You may again end up with 95% of particles in one class, in which case I would assume that all of your particles in the original dataset probably belong to the same class and the 5% just represent noise and inaccuracies.

mtherkel
Offline
Joined: 19 Apr 2015
Thanks Niko for the

Thanks Niko for the response!

I seeded the 95% class into three classes, and each of the three now has consistently 33% occupancy at various resolution limits. Should I therefore merge the classes back to one for the final refinement?

Matt

niko
Offline
Joined: 5 May 2009
Are they exactly 33% each?

Are they exactly 33% each? This would be strange.

mtherkel
Offline
Joined: 19 Apr 2015
Yes they are exactly 33.33%,

Yes they are exactly 33.33%, and have been so for 40 cycles.

niko
Offline
Joined: 5 May 2009
This means something went

This means something went wrong. I suspect that all your references are identical. How were they generated for the first cycle?

mtherkel
Offline
Joined: 19 Apr 2015
That is probably because they

That is probably because they are truly identical. I copied the .mrc from the best class to each of the three new classes and initiated refinement.

So I take it only one of the new classes should get the seeded reference, while the other two references are generated in the next cycle?

niko
Offline
Joined: 5 May 2009
All you should have to do is

All you should have to do is to increase the number of classes from 1 to 3 and update the start cycle to whatever your next cycle is. Make sure that the results from the cycle before are present. So if your refinement with a single class stopped at cycle 100, increase the number of classes to 3 and the start cycle to 101, then continue refinement by running frealign_run_refine. The scripts should do the rest.

mtherkel
Offline
Joined: 19 Apr 2015
Ok great thanks! It's working

Ok great thanks! It's working now