I’,m working on a small membrane protein with a flexible domain, I did the consensus refinement and got a good result at 3.5 angstrom resolution. However, there is a flexible small auxiliary subunit appearing only at low resolution attached to the rigid core of the protein, I did re-centering on this small region and performed signal subtraction, the local refinement still didn’t give good result. I suspect that this subunit may be only attached to a small portion of the protein given the relatively low binding affinity. I’m now trying to use 2D classification to find out particles that indeed contains this subunit (given that the heterogeneous job type does not support local search mode yet, which might be an alternative way of achieving this goal). I’m wondering if there’s any way to do 2D with limited search offset and angular search range given that this domain is already roughly aligned in the consensus refinement (something like --sigma_psi and --offset_range in relion) ?
Thanks for the request. Unfortunately there isn’t a way to do local searches in 2D classification right now. It makes sense that masked 3D classification with local searches would help in this case (if the auxiliary unit is only present on a fraction of the particles, then local refinement with the whole stack will likely not work well as the other portion of the stack will contribute only noise), and this is a fairly high-priority area that we hope to work on soon.
One alternative to masked 3D classification is using 3D variability instead. If you place a mask around the whole complex and run it with the full dataset, you might see that one or more modes captures the appearance/disappearance of density where the subunit is, which would agree with your hypothesis. Then, you could use the 3D Variability Display job to cluster particles based on that mode, and use those subsets downstream for refinements. We have a two-part tutorial on using 3D Variability that may be of interest if you want to try this out. (Links here: Part one covers the basics of the algorithm and Part two deals with ways you may display/interpret the results).
@mmclean Thank you for the detailed explanation! I see that the skip alignment 3D classification job has already been implemented in the latest version, this has already been proven useful on several datasets in our lab, we can even distinguish ligand-bound protein particles from apo-particles in some cases ! can’t wait to see how 3D classification with confined search range performs
