Using 3d classification to find loop conformations

Hi

I have a data set that consists of an icosahedral capsid bound by an antibody. When I carried out icosahedral reconstruction I found that a surface loop of the capsid had been shifted by the binding of the antibody leading to apparent disorder in the region. I did a local refinement to see if I could capture the new conformation of the loop. As part of the local refinement I did a 3d classification of my sub particles to class out any sub particles that weren’t featuring antibody binding and I was able to capture a class that represents ~20% of my particles and gives me a decent idea of where this loop shifted to. That said, I am hoping to refine this further to see if this is a hard shift to a single other conformation, several possible positions, or a more general flexible position centered around the conformation I was able to find. My current approach is to take this new map with the loop shift use it as a focus mask and repeat iterations of 3d classification to try to further class out different states for this loop. I’m wondering if anyone could tell me if this is a valid approach or if there would be a better approach that I could use? More specifically, I wanted to know if this is the right way to employ a focus mask or if I should be using the new map as an initial volume?
As an aside my initial 3d classification of my sub particles did produce a class without antibody binding that represents ~25% of my subparticles and refinements from this class show the loop in its expected conformation for the capsid.

Hi jfz002,

As far as I understand your description, I think yes, the approach is valid. You can also try masked 3DVA (same way as 3D classif) which will probably be more informative. I’m just worried about the symmetry, do you see the antibody bound to all faces of the icosahedral? If not, try to make sure you are aligning the inputs right on the binding interface. I don’t have much experience with symmetry myself, but I’m afraid it might dilute the signal if you have empty sites mixed up. Or the classifications will waste time figuring that out and might forget your loop.

Hi Carlos,

Yes so during the rounds of classifications its produced classes of unbound subparticles which I have then removed from further rounds by regrouping the classes while excluding the unbound class and then repeating 3d classification. I hadn’t thought of 3dvar, thanks for the tip.

The best of 3DVA is that it will indicate the distribution without anyone telling how many classes exist. It’ll also tell if the conformations change in a continuum, or in discrete classes. I love it.

1 Like