Difference between 3D refinement and ab-initio reconstruction

I imported particles of a small membrane protein from relion into cryosparc. I find a problem when 3D refinement, include Homegeneous, Heterogenous or Non-uniform refinement is done. There is little density detail in micelle after 3D refinement. But more detail of transmembrane helix can be seen after ab-initio reconstruction. I don’t know why. What is the algorithm difference between ab-initio reconstruction and 3D refinement? Hope any explanation or any suggestion.

They use very different approaches - see the original cryosparc paper:


Ab initio uses stochastic gradient descent, while homogeneous refinement (and NU) use branch and bound optimization.

If you get better results using ab initio, it may be worth trying going directly from ab initio to local refinement.


1 Like

Thank you very much for your suggestion. I will re-read the paper and try to do local refinement.


Have you solved your problem? I meet similar problem here, some transmembrane helix density can be seen after ab-initio reconstruction, but the helix density seems disappeared after further heterogenous, homogeneous and NU refinement. So I kind of stuck here, do not know what to do next. Hope any suggestions!

I think this problem may due to low SNR. The TM density signal is weaker than micelle. More rounds of classfication may help. However, it may does’t work for different protein.

Best wishes