I have observed several strange features on my map after NU and local refinements. The best way I can describe these are “flower-like” (tulips maybe?)
I have troubleshot several things and ruled out the following:
-These features are not due to flexible loops as they appear at structured and flexible regions of the protein.
-Adjusting mask parameters (ie. near/far, threshold) has no effect.
-Re-doing ab initio and NU refine (w/o mask applied) results in similar tulip patterns, though some will be removed and appear in other places.
Anybody see anything like this before? Any advice on whether this is cause for concern?
overfitting spikes. Often due to presence of lots of junk remaining in dataset.
3 Likes
Thanks, @olibclarke! Thought I had taken care of the junk, but I’ll see if there’s more to get rid of.
Junk, or just heterogeneity piling up along the edges of masks (if they are being used). In difficult datasets you’ll often reach a point where the challenge becomes legitimately improving some parts of the map vs. accumulating artifacts like these “tulips” in other parts. I have yet to find a silver bullet, but not for lack of trying.
1 Like
Thanks for the advice, @yoshiokc. I followed @olibclarke’s tip and removed another ~20K junk particles via hetref. Re-running the refinement with the cleaned up particle set significantly improved these spikes, though they are not completely removed, I suspect due to lingering junk as well as structural heterogeneity as you described.
2D classification can be really useful for this too. Even if you’ve moved on and classified a good bit in 3D, you can take your small subset and quickly remove some junk by 2D. Must counterbalance with angular distribution though.
@CryoEM2 I did run 2D simultaneously with hetref and found the latter to remove more junk. Though, admittedly, I did not play with the 2D settings that I probably should have so I will re-visit this to see if it can help the problem any further. I also think this could be one of the “difficult datasets” @yoshiokc mentioned.
woe is us, “difficult” ~= “interesting”
1 Like
