Sorting conformations: ab-initio or heterogeneous refinement?


I’m working on a dataset that has two clearly different conformations of the protein. After cleaning up the dataset, both ab-initio 3D with 2 classes, and heterogeneous 3D refinement with 2 classes, find the same 2 conformations. However, the same set of particles become split up:

  • ~66/33 % if ab-initio
  • ~50/50 % if heterogeneous refinement

I’m usually suspicious of evenly distributed particles because this typically means the particles are mostly randomly split up. Therefore, my first instinct is to trust more the result of the ab-initio reconstruction.
However, if I then do homogeneous refinement on the classes obtained by each method, the results from the heterogeneous refinement tend to achieve higher resolutions, contradicting my gut feeling about the particle distributions (I’m speaking here of ~6 A versus ~7 A for this particular dataset which I don’t expect to go high-resolution).
So, could it be that my conformations are really represented 50/50 in the dataset? I still doubt so, despite this giving me apparently higher resolutions… and if yes, why then the ab-initio gives a 66/33 split?

Overall both methods seem to give correct results, but I still don’t get when it’s better to use ab-initio with multiple classes, and when should I go with heterogeneous refinement… any thoughts?

Thanks in advance!