[buster-discuss] target restraints

[Clemens Vonrhein]

Hi Ed,

On Tue, Sep 18, 2018 at 03:00:18PM -0400, Edwin Pozharski wrote:
> it's entirely possible that this is a straightforward case of RTFM, please
> feel free to point it out (ideally with link to the actual manual pages
> that I need to look at).

It might as well be a missing feature - or something that we haven't
encountered in a way that would make it then into documentation or
some kind of tutorial page on the Wiki.
> 
> I am looking into using a higher resolution model to improve geometry at
> 3.7A resolution.

Good approach, yes.

> What I found so far describes what to do when the
> reference model is that of the same protein.

Indeed - that is what the normal "-target <PDB>" feature is for.

> I don't have that here, instead it's a homologue with some 40%
> identity.  From what I understood so far, buster doesn't do any
> sequence/structure alignment.

Internally it actually does check the sequence (for
homology/identity). Because it uses distances as the basis for its
local structural similarity restraints (LSSR), no superposition is
need for these cases.

> Do I have options other than mutating/editing the reference model?

Not at the moment, no.

You could restrict the targeting to main-chain only by (a) mutating
the reference model based on a sequence alignment and (b) cutting it
down to CB (or CA for Gly). This way you won't have to worry about
which rotamer to give each mutated residue - but you also "only" get a
restraint for the main-chain atoms.

> Also, I got a general feeling that this option in buster is tailored to
> cases where changes are minimal - say, low resolution data on
> protein-ligand complex of lock-and-key type.  And that with large
> conformational changes the strategy should be to include only parts of the
> reference model that stay constant.  Is that a fair characterization?

The actual LSSR restraints do have a functional form that allows for
large differences to be present: they should not have an effect on the
refinement at all (flat gradient). It is the parts that are "somewhat"
similar where one might need to do some manual pruning of the sets
under LSSR: this is what happens in the case of non-crystallographic
symmetry (NCS) when running with "-autoncs".

However, there is no feature (yet) in BUSTER to apply a similar
automatic pruning mechanism: BUSTER will take what you give it as a
target. So your characterisation is a fair assessment, yes.

What might be more useful in your case is to set up a series of
general restraints based on e.g. secondary structure - see [1] for the
underlying syntax. If you can use your high-resolution structure to
define secondary structure restraints it would be possible to keep
those active for the lower-resolution data [2].

Cheers

Clemens (for BUSTER developers)

[1] https://www.globalphasing.com/buster/manual/gelly/manual/gelly5.html

[2] We have some (not yet released) internal tools for automating the
    generation of this kind of restraints - but haven't had the
    opportunity yet to look at a significant number of real-life cases
    to assess the usefulness of this under different starting
    situations. If you/anyone has some good examples (ideally already
    published) we'd be happy to hear about them!