I ran 3DVA on a dataset with a considerable amount of heterogeneity, and it worked pretty well. I am getting a volume out of the job that is different from any coming out of classification, though, and I am wondering if this is normal or if this is most likely artifactual? My worry is that when I take this volume and use it to classify or do heterogeneous refinement, I don’t get anything that looks like this volume. It is a subtle change, so maybe only very few particles represent this, but I want to see how common it is for 3DVA to give volumes that cannot then be classified out by other means.
I think that could be normal, and it’s OK if it’s different from classification results. It might depend on how reasonable it is, considering other knowledge about your system. 3DVA could tell you about subtle, continuous motions that don’t parse well in discrete classification approaches. Also, sometimes Het Refine and 3D Classification take some fine tuning of parameters.
You could experiment a little with other 3DVA Display modes, such as Intermediate, with different bin/window sizes to take in different numbers of particles across a 3DVA mode of interest.
Remember the default 3DVA Display volumes that are usually viewed as vseries in ChimeraX aren’t technically reconstructions, if my understanding is correct, they’re volume representations of the variability. But if you run a reconstruction job on particle sets derived from Cluster or Intermediate mode, you might be able to interpret the conformational differences in the context of your system a bit more accurately.
Hi @pyoung, welcome to the forum! I’m glad you’re seeing interesting results from 3DVA!
I want to first say it’s a good instinct to be wary of things you can only see with a single processing pipeline! It’s always best to confirm that a volume really exists by trying to re-create it using other techniques. I have a few ideas of how you could do this below.
First, to mostly echo what @drichman says above, the trustworthiness of these results depends on a number of factors. If you used simple mode, the volumes you download from 3DVA Display are made by adding or subtracting the difference volumes from the consensus, so it’s possible there aren’t very many particles nearby. The intermediates and cluster modes, on the other hand, both perform 3D Reconstructions of the particles near the coordinate. So if you used simple mode, I’d highly recommend re-performing 3DVA Display with intermediates mode to see if your volume of interest still appears, both directly from the 3DVA job and from a Homogeneous Reconstruction Only job using that subset of particles.
If it does, you may want to take the subset of particles that produces that volume and perform Ab Initio Reconstruction followed by Homogeneous/Non-Uniform Refinement. The reason is that 3DVA does not change particle poses, so you may be looking at a particular subset of particles which were overfit. Performing Ab-Initio followed by a Refinement “resets” the particles’ poses, so if the volume reappears after this process you can be more confident (but still not certain!) that what you’re seeing is real.
I discuss 3DVA Display modes more in a recording from an image processing workshop here. I hope that’s helpful in getting you started, and please let us know if you have any more questions!