Hi James,

How about something not so much "blind" but near-sighted? Like adding a "missing" atom?  For example, if an ASP side chain has an OD1 but is missing OD2, I think we can all agree where OD2 actually is, regardless of density. And yet, there are PDB entries missing atoms like this.  My question is: does phenix have a tool for putting in atoms that are "missing"?

Yes. For quantum refinement with AQuaRef (https://www.biorxiv.org/content/10.1101/2024.07.21.604493v1), we need atom-complete models—seriously complete—with no missing atoms at all!

The tool for this is qr.finalise, which has been around since 2016 or 2017 (thanks Nigel!). However, it has its limitations. It won't add missing main-chain atoms, as the assumption is that if a model lacks them, it's not truly ready for fine-tuning with quantum refinement. That said, extending it to handle main-chain atoms would be trivial—we've just never needed it!

Pavel




  I don't want to play the "refmac does it" game here, but I bet not many know that refmac has a "make build Y" option.  It will do things like build in missing side chains and add an OXT at the end of chains.  Not always what you want, so not the default, but handy at times.  I have not tested how good it is at avoiding clashes or interpreting low-lying density, but it seems to do a pretty good job.  If one can add one atom at a time, then it should be pretty straightforward to do iterative "blind" building.  Perhaps using a reference structure?

But still, just wondering if phenix has a tool for adding (or at least identifying) missing atoms that I don't know about.

Cheers,

-James


On 4/1/2025 4:13 PM, Thomas Grant wrote:

Hi James,

 

A potential alternative could be to try to integrate some distance restraints into the structure prediction using other tools (Chai-1 comes to mind, but there may be others too). Then put some loose restraints just to prevent disordered regions like your N-terminus from clashing with your symmetry mates. Though, like AlphaFold, it wouldn’t be “symmetry aware” so you’d have to work out those distances yourself.

 

Tom

 

--

Thomas D. Grant, Ph.D.

Assistant Professor

Department of Structural Biology

Jacobs School of Medicine and Biomedical Sciences

University at Buffalo

955 Main Street, Room 5156

Buffalo, New York 14203

Office: 716-829-5490


From: Tom Terwilliger <[email protected]>
Sent: Tuesday, April 1, 2025 5:26:13 PM
To: James Holton <[email protected]>
Cc: PHENIX user mailing list <[email protected]>
Subject: [phenixbb] Re: for autobuild full sequence?

 

Hi James,

Oh yes I see, AlphaFold doesn't know about space group symmetry. So then you would have to...generate your partial model and all the symmetry-equivalent models that are in contact, make their residue numbers run sequentially and combine into one big model with missing residues and with gaps long enough for plausible linkers. Then supply AlphaFold with one sequence representing N full chains with the linkers in between and supply the pseudo-model.  Could work if your molecule is not too big...

All the best,

Tom T

 

On Tue, Apr 1, 2025 at 2:50PM James Holton <[email protected]> wrote:

Yes, I tried this, and found that the disordered N terminus was stabbing deep into the heart of a symmetry mate.  phenix.refine was not happy about this, and Amber went positively ballistic (so to speak).

I suppose I could start with some kind of "pre-exploded" unit cell where all the ASUs are far apart and then gradually try to bring them together, but that seems like a lot of work.

On 4/1/2025 1:26 PM, Tom Terwilliger wrote:

Hi James,

 

You could try this:  

 

1. Get your best model, including only residues assigned to sequence, with AutoBuild or whatever

2. Run AlphaFold (for example from the Phenix GUI where this is easy) supplying the full sequence and supplying your partially-built model. The resulting model should look mostly like the one you supplied, with plausible connections for the gaps.

 

All the best,

Tom T

 

 

On Tue, Apr 1, 2025 at 2:23PM James Holton <[email protected]> wrote:

Yes, but I don't want it to clash with other molecules in the unit cell, including itself.

When I was an undergrad, Steve Mayo called this an "amorphous builder".  Trivial in concept, but you need to do a "bump check" after adding each atom, and then have a plan for what to do if you hit a bump.

Make sense?

-James

On 4/1/2025 1:10 PM, Pavel Afonine wrote:

Hi James,

Are you just looking to string residues together in a line from start to end according to your sequence? That’s a quick 10-minute exercise using CCTBX, but I suspect that’s not exactly what you need.

Pavel

On 4/1/25 12:28, Tom Terwilliger wrote:

Hi James,

I think there is no way to force AutoBuild to build a full sequence when there is no density.

All the best,

Tom T

 

 

On Tue, Apr 1, 2025 at 10:26AM James Holton <[email protected]> wrote:

Hey all,

Don't worry, nothing is funny today.  I have a real question:

Is there a way to force phenix.autobuild to build in the entire
sequence?  As in: the full length of the actual molecule that is in the
crystal, such as what is supposed to go into SEQRES, regardless of
"visible" density?  I am trying to come up with a pipeline for prepping
MD simulations of protein crystals.  It seems proper to me that the
molecule being simulated should be the actual molecular species,
disordered bits an all.  However, we don't seem to have good technology
for building protein chains into "nothingness". Yes, I know Alphafold is
a thing, but it is rubbish at clashes in the context of a crystal.

I mean, I could write something, but does this tool already exist?

Cheers, and happy Tuesday,

-James Holton
MAD Scientist


_______________________________________________
phenixbb mailing list -- [email protected]
To unsubscribe send an email to [email protected]
Unsubscribe: phenixbb-leave@%(host_name)s


 

--

Thomas C Terwilliger

Laboratory Fellow, Los Alamos National Laboratory

Senior Scientist, New Mexico Consortium

100 Entrada Dr, Los Alamos, NM 87544

Tel: 505-431-0010

 



_______________________________________________
phenixbb mailing list -- [email protected]
To unsubscribe send an email to [email protected]
Unsubscribe: phenixbb-leave@%(host_name)s

 


 

--

Thomas C Terwilliger

Laboratory Fellow, Los Alamos National Laboratory

Senior Scientist, New Mexico Consortium

100 Entrada Dr, Los Alamos, NM 87544

Tel: 505-431-0010

 

 


 

--

Thomas C Terwilliger

Laboratory Fellow, Los Alamos National Laboratory

Senior Scientist, New Mexico Consortium

100 Entrada Dr, Los Alamos, NM 87544

Tel: 505-431-0010

 



_______________________________________________
phenixbb mailing list -- [email protected]
To unsubscribe send an email to [email protected]
Unsubscribe: phenixbb-leave@%(host_name)s