very odd FSC curve

Submitted by lizkellogg on Sun, 2017-03-26 23:44

Forums

Frealign

Hi,

I've been using frealign for a while now with awesome results (helical refinement for microtubule structure), but recently I collected a dataset on a new setup (krios + k2) and for some reason my FSC curve is looking very strange.

The FSC curve from my initial backprojection looks just fine, but after a round of frealign refinement followed by backprojection (14pf_r1), I get this weird feature where the FSC starts to increase again at high spatial frequencies ( < 2.6 Angstrom)
Also troubling is that the FSC curve from the refined alignment parameters does not drop to 0.

I have not changed any of the 'expert' parameters which have worked well in the past for me, I only changed the dataset specific parameters. Has anyone seen this type of behavior before and have an idea where it comes from?
Possible point of interest, I kept the res_reconstruction set to 2.6 even though my apix scaling is 0.7. Could this have possibly caused this weird artifact?

I would very much appreciate any insights or advice. Thank you!

Liz

niko Mon, 2017-03-27 09:02

This type of behavior of the FSC curve could be due to masking effects (sharp edges). It could also be due to the duplication of particles/helical sections in the dataset that produce correlations in the two half volumes used to calculate the FSC. Finally, this could be something in the data that produces strong features (e.g. ice crystals). Check your masking, make sure that particles/helical sections are not picked twice or selected twice during alignment (when particles are next to each other), and inspect your images to make sure that there are no features with high contrast.

lizkellogg Mon, 2017-03-27 15:39

Hi Niko,

thanks for your quick reply and helpful comment.. but I have to add part II of the story now.

I thought it might have been the mask on my volume, but my resulting volume does not seem to be 'cut off' in any way. I then thought it was due to over-refinement as you suggested due to duplicate particles, because I typically mask my raw particle images and for the results I just showed you I didn't.

So I went back and applied a soft-mask to my raw particle images and reprocessed my data. Now I get a FSC curve that is even weirder. My initial frealign backprojection has a striking 'dip' at 8 Angstrom resolution, clik to see strange FSC curve whereas the results from my eman2 global refinement run (from which I obtained initial alignment parameters for Frealign) show no such loss in FSC at this resolution shell. I checked my estimated defocus ranges for my data and it seems to be a homogeneous distribution of defocus which would not account for any 0's in the CTF.

I have never seen such a strange FSC curve.. any ideas what this could be due to?

niko Mon, 2017-03-27 16:03

If it is none of the problems I listed it may be more complicated. Maybe it has to do with misalignments, maybe there is a bug somewhere in Frealign. It may be difficult to resolve remotely.

Axel Tue, 2017-03-28 16:10

Hi Liz,

I have had some odd FSCs in the past when converting from Relion to Frealign (especially relion2). My solution was to do minimal initial work in relion (only 2D classification, no alignment).

Have you done your initial alignments in relion?

Axel