You are here

How to choose designed antibody

7 posts / 0 new
Last post
How to choose designed antibody
#1

Dear all,

I am trying to use Rosetta dock and antibody design modules to design antibody in order to improve the affinity of the antibody to the antigen.  Firstly I docked the antibody to the antigen. In this step I got 500 and antibody-antigen complex models and selected top 10 models based on the lowest I_sc energy. Then I ran AntibodyDesign using a command like the follow for each of the 10 complex models:

antibody_designer.linuxgccrelease \
   -s docking/ab_ag_0001.pdb \
   -primary_cdrs H3 \
   -graft_design_cdrs H3 \
   -seq_design_cdrs H1 H2 \
   -mc_optimize_dG \
   -mc_total_weight .001 \
   -mc_interface_weight .999 \
   -out:path:all design1
   -nstruct 1 | tee design.log

By doing this I get 10 designed models and the corresponding antibody sequences. But how should I select the most possible model and the antibody sequence? The score items in the output score.sc file that may be relevent include:

total_score

dG_cross
dG_cross/dSASAx100
dG_separated
dG_separated/dSASAx100
dSASA_hphobic
dSASA_int

Which score items should I use to choose the designed antibodies? 

Besides, would it be helpful to including other options for running antibody_designer such as "-do_dock", "-use_epitope_constraints",  "-paratope", etc.? And is it a routine to produce only one structure for antibody_designer?

With many thanks!

Category: 
Post Situation: 
Wed, 2020-06-24 10:06
Sunyp_IM

Hello!

 

First, I would produce many more designs than 10 for each of the complex models - if you have access to a cluster, I would definitely use that. 

Designing H3 has not yet been done with RAbD - so more models are definitely the way to go. 

For picking out models, since you are designing H3 - I would use both dG_separated (the computational binding energy), and perhaps only do this for the top X percent of total score.  Essentially, throw out the bottom 75 percent by total score; and then rank on dG_separated.  I also use the CDR instruction file to keep H3 from getting too long.  

Wed, 2020-06-24 10:12
jadolfbr

Hello jadolfbr,

Thank you for your reply. That's very helpful!

A few more questions:

(1) How many designs do you usually produce for each complex models in RAbD? 100, 1000 or even more?

(2) can the "-do_dock" option be enabled during optimizing interface energy "opt-dG"?

(3) When should the option "-paratope" be used? In the RAbD manual (https://www.rosettacommons.org/docs/latest/application_documentation/antibody/RosettaAntibodyDesign) there is a "-paratope_cdrs" option in example 3 under the "Docked design" section, but it is unrecognized by antibody.designer, I guess it is now replaced by "-paratope"?

(4) Can the stratedgy of picking up models in RAbD you described be applied in other Rosetta applications such as docking_protocol?  I mean " throw out the bottom 75 percent by total score"; and then rank on I_sc?

(5) I wish to compare the binding energy of the designs to the original complex. If the dG_separated  is the computational binding energy, how can I get this value for the original complex?

Thank again!

Wed, 2020-06-24 18:45
Sunyp_IM

1)  At least 1000.  The more rigorous your design strategy, the more models to produce.  

2) Docking will work, but opt-dG is not used during docking itself - internally, it uses it's own montecarlo and energy function for binding.

3) Yes, the docs are wrong, the option is -paratope. You would use this if you are doing docking and interested in producing models that are focused on one or more CDRs.  These are SiteConstraints, so they will guide the design to optimizing these CDRs during docking.

4) It can certainly be used for other protocols.  Here, RAbD samples the internal movement of the CDR loops and packs, so total energy can change more than some other protocols.

5) InterfaceAnalyzerMover application or RosettaScript mover. Internally, this is what RAbD uses to get the dG_separated (with the option pack_separated enabled). 

Thu, 2020-06-25 07:11
jadolfbr

Hello jadolfbr,

You commented that when doing RAbD, you use CDR  instruction file to keep H3 from getting too long. Could you tell me what is the H3 length maximun you choose? In my RAbD, I didn't restrict H3 length with the CDR instruction file, and I got many designers with H3 of 18 residues among those with the lowest dG_seperated. Is H3 of 18 residues reasonable for good designers?

Thank you.

Thu, 2020-07-23 02:56
Sunyp_IM

That's probably the max that I would use.  I would definitely caution against doing too much sequence design on it as well since we don't have profiles for those longer loops.  Perhaps express some that are sequence designed (probably using conservative design) and some that are not. 

Fri, 2020-07-24 07:24
jadolfbr

Hi everyone,

I am trying to design a nanobody using an available antibody. After grafting heavy chains from an antibody to a known nanobody scaffold, and aligning with the antibody-antigen complex structure, I obtain nanobody-antigen compelx structure. I am trying to use Rosetta design to increase its affinity to antigen as well as sample possible new CDR sequences using the options,

-graft_design_cdrs H1 H2 H3 -seq_design_cdrs H1 H2 H3 -nstruct 10000 -out:prefix graft -mc_optimize_dG -mc_total_weight .001 -mc_interface_weight .999 -do_dock -use_epitope_constraints -mintype relax -seq_design_profile_samples 5

However out of 10,000 outputs half of them contain original input sequence, with all of them having the same total score, dG seprated and all the other score parameters. Additionally, for rest of the designs the total score is high (in positive 1000s). During the run, I often keep getting the error message like this:

[FILE]: src/core/kinematics/FoldTree.cc [LINE]: 1105 [START_MESSAGE] [ ERROR ] UtilityExitException ERROR:

While the task continues to run. Is there a way to increase the sequence variability or to filter out the orginal sequences from the output?

-Thanks in advance!

Tue, 2022-08-09 20:27
chandana