I am working on a 200kDa membrane protein using DDM detergent. While I have obtained relatively good 2D classification results, the resolution after Ab-Initio, heterogeneous refinement (hetero-refine), and NU-refinement (Nu-refine) steps is only around 7Å. Here are the parameters I used:
Ab-Initio Parameters:
Number of classes: 3
Maximum resolution: 5Å
Initial resolution: 9Å
Initial minibatch size: 300
Final minibatch size: 1500
Heterogeneous Refinement Parameters:
Nice 2D classes and ugly 3D volumes are typical of flexible proteins. Are these all the particles you have? (something like 200 k?). If you can fish some more - try to get closer to 1 mi - you can play with classifications. For the moment you can check how the protein is moving, and this might give you clues about how to move on. I’d set a single homogeneous refinement with all particles from that 2D classification, and run 3DFlex on the final result, no matter what it looks like. But at this point the full resolution is not helping, so you can downsample them to apix ~3.5 beforehand.
Yes, these are all the good particles I currently have, around 200k. I think I might need to collect more data to get closer to 1 million particles as you suggested. I haven’t tried running 3DFlex yet, but I will give it a try based on your advice.
If you use a good reference a few times and do another heterogeneous refinement with the 200K good particles, what sort of resolutions do you see? Does one class dominate?
For a really small protein, I’ve recently been fighting with getting NU refinements to good resolutions also… but hetero/homo refinement gives decent(ish) maps…
Also try cleaning the 2D some more and backing off the resolution of the initial model a little; there is enough extrinsic density that I would hope a nice 10A initial model might be feasible given the wide range of views exhibited.
Thank you for your suggestions.
I have indeed done about 5-6 rounds of heterogeneous refinement, each time dividing into three classes. I usually get one relatively good class and two classes of micelles. However, the best class still only reaches around 7-8Å. Perhaps I need to continue with more rounds of heterogeneous refinement.
