You are here

Different dG_separated from RAbD and InterfaceAnalyzer

3 posts / 0 new
Last post
Different dG_separated from RAbD and InterfaceAnalyzer


I am using RAbD to design an antibody and have noticed that designed decoys will sometimes converge on the identical sequence. This seems reasonable to me. However, the dG_separated reported by "antibody_designer" in the score file is sometimes very different for these sequence-identical decoys. I wondered if maybe identical sequences were just docked differently but that does not seem to be the case as, at least in many cases, the RMSDs are 0. More importantly, "InterfaceAnalyzer" does, in fact, report identical dG_separated values for these identical decoys.

Is this expected behavior by "antibody_designer"? Should I just make it standard practice to check all decoys produced using "InterfaceAnalyzer" -- I am assuming that this is the one that I should trust more?

Here is an example of my RAbD commands:

antibody_designer.linuxgccrelease -s ${pdb_file} -graft_design_cdrs H3 L3 -seq_design_cdrs H3 L3 -light_chain lambda -nstruct 1000 -outer_cycle_rounds 35 -mc_optimize_dG -out:path:all rabd_output/ -inner_cycle_rounds 2 -cdr_instructions rabd_instructions.txt -mintype relax -do_dock -use_epitope_constraints

This produces decoys with identical sequences but reports dG_separated of -4.608 and -76.458.

The command:

InterfaceAnalyzer.linuxgccrelease -s ./*.pdb -interface HL_AB -tracer_data_print true -add_regular_scores_to_scorefile true -out:file:score_only -out:file:scorefile -inter_group_neighbors_cutoff 10

Produces the same dG_separated for both sequence-identical decoys (-17.728).

I am happy to share more details if they would be helpful. Any help would be appreciated.

Thank you,


Post Situation: 
Mon, 2022-12-19 13:58


Typically, IAM and RAbD should produce similar results.  There are a few things to keep in mind here. What version of Rosetta are you using?  There was a bug that was fixed for IAM in RAbD fairly recently. RAbD runs IAM with -pack_separated flag, which is typically the way it should be run.  For identical decoys, that does not usually happen when doing graft design and seq design unless it is outputting the started structure where the MC cycle could not find anything better.  This does happen occasionally.  You typically want to give the -random_start flat in order to start the run with random CDRs to create diversity.  If this is a very recent version of Rosetta, I can check the code for a bug, but my hunch is that it is returning the input and reporting a different score; which is still an unusual bug.  

Finally, are the input structures relaxed?   You may also want to edit the CDR instruction file to use a different sequence design strategy to increase variability or start with backrubbed structures.  

Mon, 2022-12-19 15:20

I am using the last numbered release (3.13) which I am realizing is a bit old at this point. I will update to the latest weekly release.

Thanks for the tip about -pack_separated in IAM -- I was not using that.

You were right to suggest that the MC was failing to find a better sequence. The decoys with the inconsistent scores are also identical to the starting sequence. I see how that may cause some bugginess, particularly given that I have been using an older version. Not including -random_start was kind of an experiment to see how much the sequence would end up changing -- I guess now I know.

Since you brought it up, I do have a question regarding relaxing the input structures. If the antibody and antigen structures have both been relaxed prior to docking, would you recommend also relaxing the complex once they have been docked? Or is that overkill?

Finally, I have also noticed that RAbD seems to somewhat inconsistently find/report disulfide bonds in the output pdb files. Anecdotally, it seems to find the disulfide bond in the light chain pretty dependably but fails more frequently in the heavy chain. Is this something that I should be concerned about at all?

Thanks for the help!


Tue, 2022-12-20 13:26