I am trying to use 3DVA for a ligand which I believe is flexible.
- Is 3DVA actually helpful in understanding this?
- My maps in cluster mode (during 3DVA display) are very noisy until I don’t downsample it about 4 times, which brings my effective pixel size to 2.8A and Nyquist to above 5A. But since I am looking at a ligand binding. I am looking for at least 3-4A of attainable resolution (Is this correct or would it work with that level of downsampling?). And for information I am also using a filter resolution of 3.5A since my map is at 3.2A.
Please help to find out of 3DVA is a good approach for my interest and if yes how to go about it.
Thanks.
probably not a great approach - 3DVA is measuring trajectories of motion across the whole particle. Only in a scenario where ligand-binding has a profound effect on conformation would this be a straightfroward approach. presumably you’re doing 3DVA with a very small mask? in this case I have limited insight. instead consider 3D classification, focused 3D classification with mask ~10Å around your ligand, or possibly het refine with two identical high-quality inputs and several decoy. if you have some +ligand and -ligand class you can use input mode in 3D class.
2 Likes
It really depends on the type of ligand flexibility and the protein flexibility associated with it.
Usually, it is very hard to obtain and I fully concur with the reasons and workarounds written above. (although 3DVA measures variance of voxels, so it is more precise than the whole particle).
I had a ligand that assumed several poses within the binding site or that was very mobile and I simply couldn’t refine the map correctly around it, even by following all the masking and focused refinements. For multidrug transporters for example it is often the case since there is no defined ligand-biding pocket so the ligand is free to adopt several poses.
If the ligand is however in a defined pocket, or assumes a given poses in multidrug transporters, then you can actually do it. We’ve reported here in a preprint with a detailed M&M on how to observe it: https://www.biorxiv.org/content/10.64898/2026.07.08.737148v1
The filter-resolution option is actually the extend of the search in Å, so 3Å is plenty for a ligand. But at 3.2Å resolution, be aware that you might not see the map for the ligand well. We calculated the map in simple (map deformation) or intermediate (reconstruction) modes to validate the observations. (the results are the same in both modes for this case).
Happy processing.
Vincent
1 Like
Thank you for your response.
Yes I am using a very small mask. I have tried using the same mask for 3D classification as well, But it always gives classes with equally distributed particles.
Did you try adjusting the class similarity (default almost always gives highly similar classes for me) or enabling hard classification?
Thank you Vincent.
As you mentioned at 3.2A most of the flexible region of the ligand is considered as noise by any of these algorithms.
I had a look through your preprint, Its a pretty interesting work, kudos to that. I had done something similar to that, using 12A dilated mask around the active site and 3.5A filter resolution. But the region is very noisy after 3DVA display.
Could you suggest maybe a bit more details, like how many clusters or intermediates were used during 3DVA display, if downsampling was performed to visualize the maps and how separate were the clusters in the component to component scatter plot (in my case its a uniform sphere)?
Thank you in advance.
I always have a spherical “cloud” of particle distribution in latent space, which in my hands is impossible to sub-classify using 3D classification methods (but I haven’t explored it enough to be an expert on this). It is either a continuous heterogeneity (perfect to visualize movements in 3DVA) or many sub-poses within the same general structure (which is not possible to sort out using 3DVA and challenging using other methods). I don’t downsample, and I ask for 20 intermediates (default). I have tried on 1 ligand to subclassify using cluster mode and extracting the clusters that showed the best map for the ligand to re-run refinement on; it is worth a try, it gave slightly better maps but no miracle either.
It’s hard to say more on your case without seeing more, but if your ligand is already not well observed, 3DVA is not going to help you visualize it, it is only going to blur more of it. You need to be confident of what you see before applying the method.