Manual definition of refinement alignment center

I’m not sure if this is correct way to ask but I’ll try.
Is there any way to focus the refinement center to a custom coordinate on map (with manually defining x,y,z somewhat available in cisTEM) with a goal to improve particle alignment accuracy for that specific region (I’m talking about ligand density, to clarify how small I’m thinking) to boost the map features of that particular region only ? With bigger/flexible particles that kind of focusing might cause losing overall features of the whole map but I was wondering if this is possible then it could specifically align and make ligand density look more sharp.
Please advise.


1 Like

Yes - you can use volume alignment tools for this

Thanks. But is that tool available with CS3.2 ?

Can’t remember if it was in 3.2 but it should say in the release history, it is definitely in 3.3 though

Local refinement also does it automatically, and then returns the volume to the original center at the end. By default it uses the center of mass of the mask, but you can also specify the fulcrum coordinate.

Also, just FYI, the center point in cisTEM is only for classification, not alignment. It performs 2D masking of the input images using projections of that sphere you define.

Yes although I would note that we have seen some differences using Vol Alignment tools followed by local refinement, as opposed to just using local refinement, even though as you say I think they ought to be equivalent. In a couple of cases we have got significantly better resolution with Vol Alignment tools followed by local vs local alone, even when the fulcrum is identical. :man_shrugging:

(Of course you can also use the output of Vol alignment tools for NU-refine, if you don’t want to do local refinement)

1 Like

I see…I wonder if that’s just a sampling fineness thing, or if there’s a score penalty for off-center alignments that’s not being adjusted.

I don’t know, but for really flexible/heterogeneous, peripheral domains it can make a huge difference and I don’t understand why.

For aquaporin in the ankyrin complex, we went from a messy 3.6Å consensus local refinement, to a clean 3.1Å local, and the only difference was using the outputs of volume alignment tools, no changes in the local refinement parameters. (@team maybe is this a bug? Should these two refinement strategies be equivalent?)

1 Like

@olibclarke I am very interested in trying the procedure you mention. Can you list the steps and what output you used ? I am interested in a flexible region too. Thanks a lot


  1. Make a local refinement mask for your region of interest.

  2. Measure the center of the mask (e.g. measure center #1 for the mask volume in Chimera)

  3. Input the particles and volume from a consensus refinement, the local refinement mask, and the coordinates of the mask center into a Volume Alignment Tools job.

  4. Start a local refinement with the particles, volume and mask generated by Volume Alignment Tools.

1 Like

OK, thanks for the quick reply. I’ll try and let you know. Cheers

1 Like

Hi @olibclarke. So I have tried your procedure, and got some improvements. When you do local refinement, do you have a very tight mask, or that should be rather generous ? What parameters do you think can improve things in a local refinement ? I tried going down to 3 and 2 in the angular search, but that did not make a different (got rather a bit worse). I am trying to align a few alpha helices that swing outside a larger complex. Many thanks !


  1. The smaller the region you are trying to refine, the softer the edge of the mask should be - up to 24 pixels for very small masks.
  2. Check the distribution of orientations & shifts after the first round of local refinement - if there is a large range (comparable to the initial search range, you might want to try a second round of local using the same inputs as the first, but replacing the input particles with the output from the first round of refinement.
  3. NU-regularization is on by default, but isn’t necessarily always better, so may be worth testing switching it off
  4. Using tight gaussian priors often helps in difficult cases (e.g. 3deg/1Å, with search range of 9deg/3Å)


1 Like

OK thanks, I’ll try. Best wishes

Hi @olibclarke sorry if I have not replied earlier. I have applied your pipeline, but I have a difficult case with a small protein and a lot of flexibility, so there is no magic there. In any case I am now adopting this as default as it seems to help a bit. On a different topic, I have the feeling that 3D variability works better at separating particles and corresponding reconstructions than hetero refinement. Thanks for your help ! cheers

1 Like