FSC : Relion vs CisTEM

There is a series of discussions of FSC values calculated by Relion or cisTEM at CCP-EM mailing list (click here). One of the discussion addressed by Alexis Rohou explains the details how Relion and cisTEM handles mask/unmask FSC calculations.

Here is his explanations (copied):

When comparing reconstructed volumes using FSC, one has to decide how to account for the fact that many of the real-space voxels correspond to solvent background, which is not really what we are interested in estimating the resolution of. If one just computes and FSC between two volumes with significant solvent content (it’s typical for protein to only occupy ~10% of voxels or less), the worry is that the FSC would lead to an estimated resolution much worse than that in the protein region of the map.

When comparing reconstructed volumes using FSC, one has to decide how to account for the fact that many of the real-space voxels correspond to solvent background, which is not really what we are interested in estimating the resolution of. If one just computes and FSC between two volumes with significant solvent content (it’s typical for protein to only occupy ~10% of voxels or less), the worry is that the FSC would lead to an estimated resolution much worse than that in the protein region of the map.

Relion and cisTEM follow two alternative ways to address this problem. In Relion, a mask is used which flattens the solvent (i.e. sets it to a constant value), care is taken to minimize artificial correlations due to the mask, and an experiment is performed (involving phase randomization) to try to account for any remaining “extra correlation” introduced by this mask.

In cisTEM, no mask is used. Instead, the FSC curve is computed between two unmasked volumes (well, almost.. there actually is a spherical mask to flatten the corners of the cube) and then scaled mathematically to “counter-act” the fact that many voxels are just solvent. The multiplicative factor which transforms “FSC” into “Part. FSC” is computed from the relative volume of protein within the box. 

For this reason, I think “Part. FSC” can’t be expected to be numerically identical to Relion’s mask-corrected FSC curve. However: 
– the unmasked FSCs should be directly comparable
– I am not aware of results showing that one method goes badly wrong where the other works. Generally I’d be surprised if the two methods gives very different-looking FSC curves given the same inputs

In your case, I would first check that you gave cisTEM an accurate estimate of the MW of your object. This MW estimate is used to compute the solvent correction factor. If you gave an erroneously small value for this, the scaling factor would be too large and the “Part FSC” would be artificially inflated. 

I would then do the experiment of just doing a reconstruction in cisTEM from the data imported from Relion without any refinement. You’d do this in the “Generate 3D” panel of cisTEM. This will be a kind of sanity check for whether this FSC scaling is at issue here. 

Finally, I’d add that a map at 6Å and a map at 5Å _should_ look slightly different. If you have alpha helices, for example, they should look “bumpier” / less smooth at 5Å. But this may only become apparent after sharpening of the map. Did you sharpen the maps before comparing them visually?

Hope this helps,
Alexis

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s