3DVA proper threshold at which to compare?

I’m viewing two ends (frame 0 v 19) of 3DVA simple mode output. What is the proper way to set the threshold? View them at the exact same threshold, sigma level, or check “normalize threshold” in chimera volume series viewer? I understand these can all give slightly different results, but what is proper?

It’s not mentioned in the methods paper (how they set threshold levels for figures) and the tutorial says “The ‘Normalize threshold levels’ option is nice because it adjusts the threshold for each frame so that the total volume enclosed by the mesh from each frame is the same.”

Hey @user123!

Some (delayed) thoughts:
The type of heterogeneity you expect should dictate the way you set the thresholds. Namely, if you expect compositional heterogeneity, then a fixed threshold will be more appropriate since density will ‘appear’ and ‘disappear’ throughout the sequence. For heterogeneity caused by motion/flexibility, a normalized threshold will ensure that the total density will be conserved across each ‘frame’ of the sequence. In the 3DVA paper (Punjani & Fleet, 2021), the thresholds were set for each of the target datasets given the considerations above.

Although the 3D variability components represent the variability in the dataset as a whole, a particular frame along a component axis does not necessarily correspond to individual particles observed on micrographs. We recommend that you validate the heterogeneity you deduce from the ‘simple’ 3DVA display mode in other ways, such as:

  • masked 3D classification
  • ‘intermediates’ mode 3DVA display, which will back-project intermediate volumes along a component using extant particles.

Finally, cryo-EM-based 3DVA models could also be supplemented with experiments independent of the cryo-EM data. For example, we’ve seen verification by hydrogen-deuterium exchange mass spectrometry (Josephs et al. (2021)).


Thanks for the explanation! This thresholding decision makes sense for the uses described. We had a reviewer request that the density for first and last 3DVA frames be visualized at the same sigma values, which I had not previously considered. Instead of showing the 3DVA frames, I used cluster mode and local refined the symmetry expanded particle sets and viewed these at the same sigma, which gave essentially the same result as viewing the 3DVA frames at the same threshold.

1 Like