FSC-mask auto-tightening lowers resolution in local refinement

[shuishu]

Dear All,

In the local refinement job, I noticed that FSC-mask auto-tightening makes the resolution substantially worse.

I am wondering if the output maps, in this case, are from the refinement before mask auto-tightening at 3.45 A or from after mask auto-tightening at 3.59 A. Any suggestions to improve the local refinement job are greatly appreciated.

I am using cryoSPARC v3.3.1+220118. This is an HIV Env trimer in complex with an antibody. The overall resolution is ~2.85 A. While the density for the trimer is very clear, the density for the antibody is very poor. I can see density for only the parts close to the trimer; the constant domains and part of the variable domains away from the trimer are without density. For the mask, I docked the antibody variable domains in the density and use molmap in Chimera to create a volume of the antibody variable domains. For the alignment parameters, I used standard deviation (deg) of prior over rotation 5, standard deviation (A) of prior over shifts 3, and turned on both re-center rotations each iteration and re-center shifts each iteration. For the refinement parameters, I used number of extra final passes 3, and initial lowpass resolution (A) 6. Using default parameters gave resolutions of 3.77 and 3.89 A before and after mask auto-tightening, respectively, and the output map is very noisy.

I am new to cry-EM, and this is my first attempt to do local refinement. Please give me suggestions in as much details as possible, as I am still trying to find my ways around. I greatly appreciate any suggestions to improve the local refinement results.

Best,
Shuishu

[simone97]

@shuishu

For the alignment parameters, I used standard deviation (deg) of prior over rotation 5, standard deviation (A) of prior over shifts 3, and turned on both re-center rotations each iteration and re-center shifts each iteration. For the refinement parameters, I used number of extra final passes 3, and initial lowpass resolution (A) 6. Using default parameters gave

Hi Shuishu,

I’m facing the same problem during local refinement. Have you figured out this problem yet?

Thanks,
Simone

[shuishu]

Hi Simone,

Unfortunately, I haven’t found a solution yet. I was hoping someone with experience could tell me what parameters I used were wrong, but I haven’t got any suggestions yet. I gave it up, as the local refinement didn’t improve the density at all. It could be that the region I tried to local refine was too small. It is the antibody variable region from an HIV Env-Fab complex. The antibody binding shows substantial flexibility. Any suggestions are very welcome.

Best,
Shuishu

[rposert]

Hi @shuishu and @simone97!

These are some good questions about the FSC and maks auto-tightening! I’ll try to answer them here:

Why does the quoted GSFSC resolution decrease after auto-tightening?

If you look closely at the two plots, you may note that the first plot (FSC Iteration 006) reports the Tight GSFSC resolution (3.45 Å). After auto-tightening, we calculate the Corrected curve. In this case, the Corrected curve crosses the 0.143 threshold at a lower resolution than the Tight curve, and so the reported resolution after tightening is worse (3.59 Å). But if you look at the plot key, you’ll see that the Tight curve is still 3.5 Å.

What is the corrected curve?

The corrected FSC curve is created by randomizing the phases of the Fourier components beyond a certain resolution. Ideally, this would produce maps which are uncorrelated (since, with random phases, the images should essentially be noise). However, if the mask (or anything else) has introduced overfitting between the two datasets, this random curve will still have some correlation. We can use this phase randomized curve to “correct” these artifacts.

A similar concept was suggested by Chen and colleagues in 2013, and more detail on CryoSPARC’s implementation is available in our guide.

Why is the corrected curve worse than the tight curve?

The fact that the corrected curve never rejoins the tight curve indicates that the dynamic mask is indeed too tight — you may wish to provide a static mask with more dilation or padding. We have more guidance on mask creation in the guide.

Local refinement

Local refinement is a very useful tool, but as you’re seeing, it can be challenging to apply to small regions. @shuishu, in your specific case of a small antibody region, the local refinement may benefit if you first perform Particle Subtraction. This will (ideally) remove the signal from the part of the protein you are not trying to align, which can make it easier to align the remaining signal from the antibody.

We have an extensive tutorial on Local Refinement which covers Particle Subtraction and other topics — reading through that may give you some other ideas too!

---

I hope that’s helpful! Please do post again if you have any further questions on the FSC curves created by CryoSPARC or Local Refinement!

[shuishu]

Thanks very much for the explanation. That is very helpful.

Best,
Shuishu