Particles (and mics) imported from CS to Relion are very noisy

Dear all,

I imported particles refined in CS to Relion using I exported my consensus refinement and provided its output …_particles.cs file as input. Then did essentially the modifications instructed in the pyem wiki page. The only differences as far as I can tell are that (1) for some reason pyem included additional “>” symbols in the resulting .star file that I had to remove (using sed -i “s/>//g” < filename>) and (2) I had to manually add_rlnOpticsGroup into the resulting .star file. Without this step, I get errors on import. After import, the particles as well as the micrographs look very noisy, but for example I can see edges of holes in the micrographs. This tells me that they are in some ways “correct”, but maybe the normalization is somehow off. If I reconstruct these particles, I get a map that matches the one in CS. However, if I try to do subtract projection, I get a negative density of the parts I want to keep in the remaining particles, which again points (in my mind) to an error in normalization.

Is there something different happening in terms of normalization of particle images in CS compared to relion or is there something wrong with my approach using pyem?


It sounds like you have an old pyem - I suggest updating. Aside from that, particle normalization is only done on extraction (in either program). You need only do a new reconstruction in Relion, so that the map is on the raw image intensities. Relion doesn’t know about cryoSPARC’s scale parameter (alpha), it needs a fresh map from relion_reconstruct.

The current version of Relion uses a “mask what to keep” approach to subtraction, but old versions were the opposite. It sounds like you may have this part backwards, since what you want to keep is negative - meaning it was actually subtracted.