I have recently encountered an interesting case where the quality of my map after refinement is significantly better than what is reported by the corrected FSC (the map is showing typical features of around 4-4.5 Å resolution, but corrected FSC has a very sudden, sharp drop at ~7.5 Å, dropping to even considerably negative values. The tight FSC curve does not display such feature interestingly).
I fear that because of the fact that the corrected FSC sets the maximum alignment resolution, I am not reaching the best possible reconstruction. So I first tried to disable the high-resolution noise substitution during FSC computation, but there seems to be a bug that prevents this option from being turned off. Submitting a job through cryosparc tools setting this parameter to False seems to not work either, since in the metadata of a job created in this way, the value for the parameter is still set to True.
So short of it, I tried to set manually the maximum alignment resolution to 3.5 Å through option Maximum align resolution (Å).
However, checking the job log, at each iteration it still reports “Using Max Alignment Radius 7.0000Å”.
Therefore I am wondering if there would be any way to either disable high-resolution noise substitution during FSC calculation, or to bypass the maximum alignment resolution properly?
Thanks a lot in advance!
My understanding is that the maximum alignment resolution sets the maximum resolution used for alignment, but otherwise (below that cutoff) it will still be set by the FSC.
Which refinement job type is this? If it is local refinement, you might consider using a softer refinement mask.
Thanks a lot for your reply!
This is in fact from a helical refinement job.
I see! So this parameter works opposite as to what I thought, instead preventing the usage of higher resolution features for alignment.
I guess in this case the only way to test if the reconstruction can get better without the limitation in resolution used for alignment imposed by the corrected FSC would be to just disable the option to perform high-resolution noise substitution… but this option seems buggy currently
As a side note, I do in fact get an estimate of resolution closer to what the map looks like (~4.5-5 A) by performing symmetry expansion and then doing a local refinement using a mask that covers just the 2 most central helical turns.
Optimising helical refinement is still quite tricky and getting masking and helical parameters right can be a process of iterative experimentation. Worse, it’s really easy to settle into local minima which “look” right (both visually and by metrics like FSC) but are misleading… then identifying the correct ones and “flexible” or partial occupancy features suddenly snap in much more clearly.
Please can you post the FSC curves that you mentioned in this post.
We will release a fix that will allow disabling FSC-based noise substitution, but the fix will unfortunately not quite make it for inclusion in v4.3.
Sure! Please find here an example:
Note that this is with data binned 2x.
Even though the corrected FSC is reported to be 8.5 Å, the map shows features such as side chain density and β-strand separation, more typical of a ~4.5-5 Å resolution in my experience.
I have found a few more interesting observations that I thought might be worth sharing. Firstly, the very sharp drop in corrected FSC seems to happen at more or less the same point in the range of FSC values (around half-two thirds of Nyquist frequency) regardless of the binning level. See below an example with data binned 3x, but I have also observed the same with the same data binned 4x:
Additionally, I have realized that the resulting maps contain highly negative regions, with negative values almost up to the same absolute value as the highest positive values. See here an example of intensity scale:
I tried to run the refinements with enforce non-negativity. Indeed, by doing so I get a much better FSC with noise substitution, but the map quality is considerably worse.
I am a bit at a loss here! Have not found a case like this before. Any insights would be greatly appreciated!