Hi,
Following the standard protocol I ran an ab-initio reconstruction of a dataset with one class. This generated a 12Å model, which I used in refinement. Refinement finished with a final structure at 17A - way worse than the original input from ab-initio reconstruction.
The same happened for two other datasets - ab-initio models look more featureful and have higher resolution then the results of refinement.
Strange… how many particles in the datasets? Have you processed these before in other programs/what kinds of results did you get?
One thing to note is that in ab-initio reconstruction, the resolution number is not actually a resolution estimate, it’s an indicator of the maximum resolution that’s being considered by the optimization routine. Though what you’re reporting is still very odd, especially if the ab-initio models by eye look better.
Dataset has ~100k particles with 2.5 Å pixel size. Processing with other programs gives a very low resolution blob of the correct shape, so cryoSPARC results are way better.
Classification with 3 classes and independent refinement stops at ~17-18 Å for all of them. One thing that gave better results was turning on dynamic masking at 20 Å (default=12). This gave me a 10 Å final refined map. At this point my limitation is probably due to the data itself.
Attached: comparison of ab-initio and refinement with default settings and refinement result with dynamic masking starting at 20 Å.
–
