Unrealistic rotamers for D-amino acids with aromatic side chains. Side chains appear to be skewed/smeared in output pdb.

[tsamdin]

I'm getting unrealistic rotamers for D-aromatic side chains. I made sure to include the LimitAromaChi2 task operation - but I'm not sure if this applies to rotamers for D-amino acids? The rings appear skewed/smeared - with unrealistic bond lengths?

Rotamers for L-aromatic side chains in the rest of the protein (not shown) appear to be realistic, so I believe the issue is limited to just the D-amino acids.

Any recommendations on how I can address this behavior?

[everyday847]

Some observations:

• The D-amino acids are constructed by an exact mirror image flip of the L-amino acids. All the ideal bond lengths and angles are identical, up to PDB rounding errors.
• The rotamer libraries are similarly flipped, and only have to do with dihedral angle values.

So, two questions:

• Are you packing in the context of relax? Are you certain this is not happening during the minimization steps?
• Are you running a non-ideal relax variant (i.e., either letting bond lengths and angles vary in internal coordinate relax, or an explicitly Cartesian relax)?

If so, my highest suspicion is that this is an issue in cart_bonded where some parameters (particularly for rings) are not cloned to D-amino acids.

[tsamdin]

I'm going through my XML script and looking more closely at my relax/design/pack movers. One thing that stands out to me immediately is how I set up my MinPackMover. I had nonideal set to 1 - which from the documentation allows "nonideal (minimization) : open up the bond-angle- and bond-length DOFs to minimization". I'm going to set nonideal="0" and see if that makes a difference.

<MinPackMover name="MPM" scorefxn="r15_cart" nonideal="1" off_rotamer_pack="1"
task_operations="use_input_rotamer,extrarot,no_repack_L_protein,no_design_L_protein,L_only_design,D_only_design,D_buried,L_buried,D_strand_buried_resnum,D_strand_exposed_resnum,no_cys,no_Gly_CYS_DCYS,conservative,no_ASP_DASP,aromachi"
/>

[tsamdin]

Setting nonideal=0 did not prevent the generation of unrealistic D-amino acids with rings.

I found a link to an old forum post that describes a similar issue https://forum.rosettacommons.org/node/3620. I was using ref2015_cart_cst.wts for the score fxn used by MinPackMover, so I'll try switching to the recommended score12minpack.wts. - hopefully this is the fix.

Also in my setup of the MinPackMover, I don't have cartesian turned on for MinPackMover, so I'll give that a go as well.

[everyday847]

Yeah, I'd systematically take movers out and add them back in to narrow down which component we need to dig deeper on. Anything that minimizes is suspect. (Similarly, any protocol that sets up a scorefunction or minmover using some unusual logic -- if it ignores some input flag it should consider or whatever.)

[vmullig]

To check whether this is a packing issue or a minimizing issue, try replacing the MinPackMover with the PackRotamersMover. The output should just be on-rotamer conformations then.

…

On Tue, Nov 19, 2024 at 1:53 PM Andy Watkins ***@***.***> wrote: Yeah, I'd systematically take movers out and add them back in to narrow down which component we need to dig deeper on. Anything that minimizes is suspect. (Similarly, any protocol that sets up a scorefunction or minmover using some unusual logic -- if it ignores some input flag it should consider or whatever.) — Reply to this email directly, view it on GitHub <#245 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABACZUBYG6NASIZWMCRQQCD2BOCLNAVCNFSM6AAAAABSCIDLHCVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCMZRGAYDSOI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.*** com>

[tsamdin]

So I tracked down the issue to a Fastrelax mover that followed minpackmover, where I had bondlengths and bondangles=1. I wasn't using dualspace or cartesian in this mover.

What I did to resolve my problem was turn on dualspace, and change the score function to one that was setup for cartesian minimization. As a follow up question, should I also add 2 rounds of cartesian minimization as described in the relax documentation?

Use the Dualspace protocol for dihedral and cartesian minimization as described by Conway et al. Do 3 FastRelax cycles of internal coordinate relax followed by two cycles of Cartesian relax - cart_bonded energy term is required, pro_close energy term should be turned off, and use of -relax::minimize_bond_angles is recommended. The -nonideal flag can be used for this.

https://docs.rosettacommons.org/docs/latest/application_documentation/structure_prediction/relax

[everyday847]

You don't have to, but you can. That's the status of innumerable protocols. That paper showed better results on a benchmark that may or may not be related to your scientific use case.

…

On Sat, Nov 23, 2024, 8:17 AM tsamdin ***@***.***> wrote: So I tracked down the issue to a Fastrelax mover that followed minpackmover, where I had bondlengths and bondangles=1. I wasn't using dualspace or cartesian in this mover. What I did to resolve my problem was turn on dualspace, and change the score function to one that was setup for cartesian minimization. As a follow up question, should I also add 2 rounds of cartesian minimization as described in the relax documentation? *Use the Dualspace protocol for dihedral and cartesian minimization as described by Conway et al. Do 3 FastRelax cycles of internal coordinate relax followed by two cycles of Cartesian relax - cart_bonded energy term is required, pro_close energy term should be turned off, and use of -relax::minimize_bond_angles is recommended. The -nonideal flag can be used for this.* https://docs.rosettacommons.org/docs/latest/application_documentation/structure_prediction/relax <http://url> — Reply to this email directly, view it on GitHub <#245 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABGGY26ZAHYTEICDLQHWF4L2CB6AFAVCNFSM6AAAAABSCIDLHCVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCMZVG4YTMMY> . You are receiving this because you commented.Message ID: ***@***.***>

[tsamdin]

Thank you!

[roccomoretti]

What I did to resolve my problem was turn on dualspace, and change the score function to one that was setup for cartesian minimization.

It was the changing of the scorefunction which did it. You can do flexible bond lengths and angles with the regular (atom tree) minimizer/relax, but since the regular scorefunctions don't contain any terms which restrain those DOFs, you can get messes from it. The cart_bonded term adds restraints for bond lengths and angles. That's very important for Cartesian (and dualspace) minimization, but it's also needed for flexible bond length and angle minimization in internal coordinate space as well. (So 'cart' here is a bit of a historical misnomer.)

[tsamdin]

Thank you!