You are here

Antibody protocol: unusual antibodies, and chainbreaks

6 posts / 0 new
Last post
Antibody protocol: unusual antibodies, and chainbreaks
#1

Hello,
I am doing a research internship, on modelling some antibodies recognizing a short peptide from a conserved part of the gp41 protein of the HIV1 virus and simulating their docking.
I have been successfully using the ROSIE Antibody app, but I have had some sequencing issues (the samples were contaminated, and several rounds of PCR were used) . On a couple of different antibodies, I get a L1 loop ending in WYL rather than more usual residues such as WSL. I have not found similar antibodies in databases, and this triggers an error from the IdentifyCDRs method of the module antibody.py: ROSIE won't accept such a sequence. I would like to get a similar output to the one of the Antibody server, since I will next be doing some docking and comparing various antibodies targeting the same epitope.

I downloaded Rosetta (2014.22.56873 bundle) and compiled it successfully. In the rest of this post, I am writing about the antibody protocol as found in tools/antibody. I just changed a line in IdentifyCDRs so the WYL ending would be accepted.
# L1
res = re.search( r'C[A-Z]{1,17}(WYL|WSL|WLQ|WFQ|WYQ|WYH|WVQ|WVR|WWQ|WVK|WYR|WLL|WFL|WVF|WIQ|WYR|WNQ|WHL|WHQ|WYM|WYY)', light_first)

When I run the antibody.py script, specifying the rosetta binaries, the part about grafting antibody CDR loops fails due to loop modelling issues. This happens even with example sequences. The antibody protocol fails when the python module attempts to move the missing PDB file that is the output of the CDR grafting. Log files shows that there is an error at 100H, away from WYL. Here is an example:
protocols.loops.util: checking peptide bond geometry:
protocols.loops.util: 100 H C-N 517.305 max: 2

So, here are my questions:
- How can get an output similar to the Rosetta Antibody Server for unusual antibodies? I need to be able to compare them with previous antibody jobs (that went fine). I would like to get several models, using the same scoring function.
- Why do I keep having these CDR loops grafting issues? Is my install wrong or should I run another protocol to close these loops?

I am posting a few files containing the antibody protocol output (using the verbose flag), for one of my unusual antibody and for sequences given as examples. I also joined the self test, which fails for some antibodies(?), but this seems to be a numbering scheme issue. My antibodies are detected correctly after the change to IdentifyCDRs. I ran the unit tests after compiling Rosetta, and I do not recall having had any errors.

I am going to try to run the Rosetta Antibody Server on antibodies with a corrected L1 ending (WSL) to make some comparisons and see if it really impacts my structures. Having two different antibodies ending in WYL means it might not be a problem with the sequencing.

Any help will be appreciated, thanks for this great software and the server version of it!
Cyril Cros

Post Situation: 
Mon, 2014-07-21 17:03
Cyril Cros

I realize this post has been unanswered for almost 2 months. I am having the same issue with FR_0001.pdb not being found and the antibody model protocol failing. This should be fixed soon. Sorry again for the late response.

Mon, 2014-09-29 11:08
jadolfbr

Actually, the error you are seeing where it fails about the geometry has been fixed for quite some time. Have you ran it again recently?

Mon, 2014-09-29 11:12
jadolfbr

Hi, thanks a lot for the answer, I guess I am not quick to reply either. I just looked back at my issues.
I solved some problems by using the "-check_cdr_chainbreaks false" flag for the loop modelling. There was actually no break in my structure...
I have since moved on to the newest Rosetta version (September 2014), on a new computer also (with Profit 3.1 and the NCBI+ blastp installed as well). I now get a new kind of complaint from the job distributor, about a residue not being a cut-point variant. I only edited the antibody.py arguments to point to the correct paths, and to remove the -q argument (for quick ie no relaxation). See attached files, with the input files and the result I got in verbose mode. It is the same with the sample data (VL5 and VH5). I also still fail the self-test...
I used Linux (Archlinux), kernel 3.16 and gcc 4.9 for the compilation, python2.7 for the script.
Do you have any information on this old bug? I used the 2014.357232 bundle, the protocol is in tools/antibody. It has not changed a lot since the last time, I think.
I am also hearing (cf "Blind prediction performance of RosettaAntibody 3.0: grafting, relaxation, kinematic loop modeling, and full CDR optimization" Proteins 2014) that a new version might come out. Any news on the release date?
Thanks a lot for your help!

File attachments: 
Mon, 2014-11-24 08:16
Cyril Cros

This is also failing in the developers version. Running locally, it fails at the same spot and has been failing for quite a while. I have contacted the author, hopefully he will be able to fix it shortly.

Tue, 2014-11-25 08:02
jadolfbr

Thanks a lot for the update!

Tue, 2014-11-25 08:21
Cyril Cros