Hi all,
A few questions that I think might all relate to the same subject:
We have a map and the resolution is currently 3.0A, but given the quality of the data, number of particles and how the map looks like, it seems to rich higher resolution- only that we’re stuck at 3.0A.
When applying the defocus refinment in the homogenous/ NU refinment, the resolution improved f
First, I tried to global CTF refine followed by local CTF refine, however the graphs, when done in med resolution (5.0A) seem mostly like noise and lack local minima:
Does anyone have an idea how to improve it?
One of the ways to solve this, I generated optic groups, and got 76 (makes sense given the way the data was collected). See the parameters-
However, when I simply placed the output particles into Global CTF, I got “group 1 of 1”. Are the particles split into optic groups, or do I need to change something in the parameters?
Thanks, here it is-
I first tried to do global first, then local CTF.
Later, I found a recommendation online to do the opposite.
However, regardless of the order, the 5-8A Min fit in local CTF results with failure to identify real minima, and the 20A min still doesn’t look very good.
I hope the tree is clear?
Hi @Amitm22! We’re still trying to figure this one out, and I have a couple more questions for you.
Can I ask why you’re changing the maximum resolution parameter? Generally we recommend this is left as the default, which will use the GSFSC resolution.
Also, is there a reason you’re using such a small pixel size (i.e., why you’re not Fourier-cropping your particles)?
Hi,
Sorry for the mistake - I meant the Min resolution, which is recommended to be adjusted to mid-high res (according to Cryosparc’s little “info”).
I leave the Max untouched.
After several rounds of cleaning the particles in 2D (binned ~6 times) I re-extracted to the original pxl size. Would you suggest doing it in the binned size? Could you please explain the rationale?
Thanks
Thanks for the clarification! That might explain why your local CTF refinements look strange. Local CTF Refinement uses only information in the range from the Min to the Max resolution. In your case, your Max resolution is the GSFSC (approx 3 Å). So when you set the Min resolution to 5 Å, the job was using a very small amount of information (from 5 to 3 Å) to fit the defocus. Leaving as the default (20 Å) lets it use much more information, so there’s just the one local minimum you can see in your J76.
You indicated that the fit for J76 (the 20 Å minimum resolution job) didn’t look good to you – could explain more what you’re looking at and why it looks wrong?
Pixel sizes
Generally, there are two competing forces to consider when extracting particles:
Your map cannot be higher resolution than twice your pixel size (this is called the Nyquist limit)
Smaller box sizes (equivalently, larger pixel sizes) process significantly faster
Because of point (2), you want to use a box that’s as small as possible to make jobs faster. However, because of point (1) you want your pixel size to be at least half your best GSFSC resolution. In practice, people generally recommend setting your pixel size such that your GSFSC resolution is approximately 3/2 your Nyquist limit. For example, if your maps are generally around 3 Å, a pixel size of 1 Å is a good starting place.
In your case, it looks like your pixels are very small. I can’t quite see in the image you posted, but I’d guess they might be around 0.4 Å/pix? If so, your Nyquist is 0.8 Å/pix, but the particles are only refining to around 3 Å! You could save a lot of processing time using Downsample Particles to make a new particle stack with a pixel size of 1 Å with no lack in quality.
Thanks…
From what I read online, a good minima from local CTF should be a few hundreds drop, whereas, in this case, even the 20A, gave only around ~15 units drop (see below) - what do you think?
From the local CTF tutorial: “Therefore, plots with a minimum that is hundreds of units deep indicate that we are highly confident about the optimal defocus value. On the other hand, plots with very shallow minima (tens of units) indicate uncertainty in the optimal defocus.”*
As for resolution, thanks for the explanation. With this number of particles and micrographs, and 2D that looked very good (with different orientations), we were hoping to achieve ~2-2.5A resolution. Even imposing a C2 symmetry (the particle is a homodimer) didn’t improve the resolution in more than ~0.1A.
I can try to re-extract to a lower pxl size, say 0.8A/pxl, about the 1/3 of the resolution we’re aiming for.
Ah, great point – sorry I missed that! You’re right that a shallow minimum for Local CTF Refinement means we’re not certain about the best defocus for that particle. However, that uncertainty is not necessarily due to something being wrong with the Local CTF Refinement. It could be because of problems with particle poses, poor resolution, a bad or noisy volume, low signal-to-noise ratio in particle images, etc.
If there’s only one minimum but it’s broad and shallow, it may be best to look at earlier steps in the processing pipeline to see if you can improve these other factors which affect confidence rather than parameters in the CTF refinement itself.
I also forgot to ask about the exposure groups. If I’m understanding correctly, you:
Plugged exposures and particles into an Exposure Group Utilities job (J71)
Set Input Selection to exposure
Split the exposures into 76 exposure groups
Plugged the output of J71 into a Global CTF Refinement
If so, I think what’s happened is that since the Input Selection parameter was set to exposure, the particles did not have their exposure groups set. What happens if you clone J71 and turn on the Correspond particles... parameter (see below), then try a new Global CTF Refinement with those particles?
Do you have some ballpark figures for more and less certain defocus accuracy? I’ve usually worked on whether or not there are multiple potential minima within the defocus range for how trustworthy an estimate is, rather than the depth (for example, if a large defocus range but one dip, I’m more likely not to worry too much versus two or three potential minima visible)…
When I perform CTF Refinements I operate in this way as well – if I see one minimum I don’t worry too much about how deep it is. I think I’d worry more if I saw a shallow minimum (10 – 40 units) that was far away from 0 (i.e., the whole-mic average defocus); put another way, I’d worry if I was significantly changing a particle’s defocus based on low-confidence information. But I don’t have anything other than my intuition to back that up.
Thanks for the suggestion.
When i did that (unticl Correspond particles…) initially, the global CFT assigned all particles to one group instead of 76. I had to manually change it to + to get the “by optic groups” sorted.
So, initially I tried: optic groups -->global refinment → loacal refinement,
Then changed the order:
optic groups → local ctf (at 20A) → global refinement (at 8A).
I do see a difference if I change the order (global or local first), but frankly, still struggling to understand the meaning of it and how to further filter the less good groups.
Here are two examples for the same group, when done global first:
I would say that anecdotally, we very rarely see minima that are 100s of units deep, even when per particle defocus refinement has clearly “worked” as judged by improvements in resolution and map quality.
I do wonder whether it might be possible to calculate a better (dataset level) measure for fit confidence, rather than just eyeballing based on the first few individual particle plots. (Also perhaps a threshold for fit confidence or estimated defocus change beyond which to reject the change in defocus and just use the original value…)
It would also be helpful to have defocus variation mapped on the micrograph (as relion does in CTFrefine), so we can see if there are any patterns, or any obvious outliers (e.g. particles near gold/ice).
Thanks!
The thing is, I had in the past datasets that were by far less good than this one, but the local CTF came back with stronger minima values (drop of tens, not ~15, units).
So I wasn’t sure what to make out of it.
Also, can someone please recommend the next steps to improve the data? I found the tutorial a bit insufficient…
Thanks!
That is what I meant - thanks. It is hard to judge much from the fits without seeing the data going in, and the residuals coming out.
At this resolution, I would be surprised if correcting Cs/tetrafoil makes any appreciable difference to resolution & map quality. 0.7mrad beam tilt definitely could, though.
Also, Cs is refining well away from 2.7mm, which could indicate a slightly off initial pixel size (or, more likely), inaccurate fitting of this parameter.
Where would you change the beam tilt to 0.7?
Regarding the pxl size: We have a pdb of this particle in a different conformation, and the maps look great - we get a very nice fit, the helices seem in the right size, etc.
We had previously data where we had to re-adjust pxl size slightly, and it didn’t fit the map as good as this one. But thanks, I’ll check with our microscopist about the Cs.
No I wasn’t suggesting changing the beam tilt - I was noticing from your log file that it was refining to ~0.7, and saying that that will likely make a difference at 3Å, where as tetrafoil & Cs are unlikely to matter at this res unless very severe (so I wouldn’t bother refining them).
