Hello,
I’m trying to use 2D classification to separate viral particles with and without DNA, so that I can assess their population. However, I find that no matter what I try, the particle set of viral particles with DNA always contains a significant amount of viral particles without DNA. I assume this is because there is just too much alignment in the envelope between full and empty particles that the random orientation of DNA content just becomes noise and is disregarded during 2D classification.
Does anyone have any suggestions how I could achieve my initial goal?
Thanks in advance!
In my experiences of trying the same thing, it’s nearly impossible to separate full and empty via computation purification. There will always be some “full” particles in the “empty” dataset, and the reverse. I had some success with subtracting the capsid and classifying with alignments off, but it was still not perfect and was extremely computationally expensive.
As long as the dataset isn’t millions of particles, manual eyes-on is still the best way, IMO. If you’re a dab hand with machine learning, you could try writing a sorter…? See if that works better.
Thanks for replying. Do you know the easiest way for me to create a gallery of all extracted particles so we can go through them by eye? I think we could maybe sort them based on intensity and then count through. Cheers.
Not sure, the times I’ve had to fall back to that, I’ve done it in RELION (maybe pyem with csparc2star.py on the particle stack, then relion_display will serve?)
Well I think I’ve seen advertising of one of the 3D classification methods exactly on this kind of problem. Just not sure which one it was (if 3DVA, 3D classification, hetero refinement…). Of course you’ll need to tweak the parameters, no matter which one you chose. And the more heterogeneity you have, more particles you’ll need, no secrets here.
Have you attemped to used 3D-classification with a solvent mask around your capsid and a focus mask around the interior contents? It might also be worth generating a volume with no contents in the center and a volume with contents (using symmetry relaxation in a 3D homogenous or NU-refinement), and then using these volumes in a heterogenous 3D refinement.