I can definitely point you in the right direction @XueyinZhou! I’ll provide more detail below, but based on the images you posted, I would expect that you’d see an improvement in the regions you indicate by using ab initio reconstruction and heterogeneous refinement to ensure you have a clean particle stack before proceeding with other techniques.
For particle curation in general, we would recommend following a procedure where 2D Classification is used to remove the most obvious junk picks, then ab initio reconstruction and heterogeneous refinement are used to clean up the particle stack in 3D. I have outlined this process here. Your case is obviously different from that one, since you do not have oligomerization to deal with, but the general procedure will remain the same.
Once you have a clean particle stack (that is, heterogeneous refinement generally shows you copies of the same class), you should try performing a homogeneous or non-uniform refinement. This is commonly called your “consensus refinement”.
If your GSFSC resolution here is better than 5 Å, we typically recommend trying 3D Classification, 3DVA, or 3D Flex to tease apart conformational differences. If it is worse than 5 Å, we typically recommend that more particle curation is performed, since it is likely there are still contaminating particles unrelated to
your target in the particle stack.
I hope that helps!