Hello everyone,
I am processing data for a flexible protein complex with RNA, total size around 125 kDa. Data was collected on K3 with an energy filter, processing in Cryosparc v4.7.1.
I used template picker (generated by using AF prediction volume and Topaz, then after getting an initial volume I generated templates from it and used them to pick on the whole dataset). Blob picker has been giving me nothing but noise (visually picking in between what I consider particles).
After multiple rounds of template-picked 2D classification (aiming for <10k particles per class, circular mask diameter 150 Å (an informed guess), number of final full iterations 2, number of online-EM iterations 40, batchsize per class 200), I got several views of the particles with sharper features per class (4-6 Å).
Used HR-HAIR for ab-initio (3 classes, initial minibatch size 300, final minibatch size 1000, maximum resolution 5 Å, initial resolution 8 Å, Fourier radius step 0.005, center structures in real space disabled) and 200k particles from these 2D classes. One of the ab initio volumes looked promising as to what I am roughly expecting to see.
However, when I try to use NU/hetero refinement from single or multiple ab initio volumes, the quality of the structure goes down, keeping the expected shape but becoming more “hollow“ in the middle, showing resolutions in the range 5-6 Å, but virtually less structural features. 3D classification does not produce distinguishable classes.
In general, all of the multiple ab-initio runs, hetero refinement runs, 3D classification, etc, just evenly divide the particle stacks into the number of input volumes, without differentiating between lower and higher resolution data. In addition, those classes where RNA is clearly present (more contrast), the protein density seems blurred and likely interfering with the alignment.
Any thoughts and advice on the 3D reconstruction would be greatly appreciated (I am a newbie in Cryo-EM data processing). I am not sure if this is particle picking issue, heterogeneity/flexibility of the complex, RNA brightness interfering with the reconstruction or maybe over-fitting during refinement (so far for refinements I used dynamic mask)?
Thank you very much in advance!