Author: Jianqing Xu (, Daisuke Kuroda (, Oana Lungu (, Jeffrey Gray (

Corresponding PI Jeffrey Gray (

Last edited 5/23/2017 by Jeliazko Jeliazkov (


We recommend the following articles for further studies of RosettaAntibody methodology and applications:

  • B. D. Weitzner*, J. R. Jeliazkov*, S. Lyskov*, N. M. Marze, D. Kuroda, R. Frick, J. Adolf-Bryfogle, N. Biswas, R. L. Dunbrack Jr., and J. J. Gray, "Modeling and docking of antibody structures with Rosetta." Nature Protocols 12, 401–416 (2017)

  • B. D. Weitzner, D. Kuroda, N. M. Marze, J. Xu & J. J. Gray, "Blind prediction performance of RosettaAntibody 3.0: Grafting, relaxation, kinematic loop modeling, and full CDR optimization." Proteins 82(8), 1611–1623 (2014)

  • A. Sivasubramanian,* A. Sircar,* S. Chaudhury & J. J. Gray, "Toward high-resolution homology modeling of antibody Fv regions and application to antibody-antigen docking," Proteins 74(2), 497–514 (2009)


Please realize this the overview is to speed you up to run the protocol asap with minimum knowledge. For details of each steps, please check:

Rosetta Antibody can model both antibodies (consisting of the heavy and light chain variable region) and nanobodies (consisting of only the heavy chain variable region). To run the protocol, one needs:

  1. The sequence of interest in FASTA format, with a description lines preceding the sequence and indicating either ">heavy" or ">light".
  2. BLAST+ (version 2.2.28 or later).
  3. Rosetta (the latest if possible, officially supported in 3.7).
  4. The antibody database contained in the Rosetta/tools repository (as up-to-date as possible).

In Rosetta, antibody modeling is a two stage process.

Executable: antibody

First, the sequences are divided into structurally conserved regions (FRH, H1, H2, H3, FRL, L1, L2, and L3) and templates are selected from the database based on BLAST+ score. Alternatively, manual templates can be specified via PDB code and the -antibody:{region}_template, for example: -antibody:l1_template 1rzi. Note that the templates must be present in the database. After selection, template complementarity determining regions are grafted on the template frameworks and the frameworks are assembled according to a template VH–VL orientation (predicting this orientation is challenging, so ten template orientations used). A highly recommended, but optional, FastRelax with constraints is used to alleviate any clashes introduced by grafting.

Sample command line: antibody.macosclangrelease -fasta antibody_chains.fasta | tee grafting.log.

Here antibody_chains.fasta looks like:


The typical runtime, with FastRelax, is ~20 mins per model or ~3 hours for 10 models.

Executable: antibody_h3

Next, for each grafted model, the CDR H3 is de novo modeled and the relative VH–VL orientation is refined via local docking. Flags are split into a simulation set and a loop-modeling set. Both sets of flags are shown below (the loop modeling flags can also be found in tools/antibody/abH3.flags). If loop modeling is slow, it can be expedited by decreasing KIC sampling via the flags -loops:refine_outer_cycles 2 and -loops:max_inner_cycles 20, however these flags have not been benchmarked. If using multiple VH–VL orientations, we recommend 1000 structures be generated for the top grafted model (typically, model-0.relaxed.pdb) and 200 structures be generated for the other orientations.

Sample command line: antibody_H3.macosclangrelease @flags.


# input grafted model
-s grafting/model-0.relaxed.pdb

# recommended number of structs
-nstruct 1000 

# recommended kink cst, kink present in 90% of Abs

# necessary if running multiple procs w/o MPI

# specify output file
-out:file:scorefile H3_modeling_scores.fasc 

# specify output folder
-out:path:pdb H3_modeling 



#how to run antibody mode -- these are the current best-practices
-antibody::remodel              perturb_kic
-antibody::snugfit              true
-antibody::refine               refine_kic
-antibody::cter_insert          false
-antibody::flank_residue_min    true
-antibody::bad_nter             false
-antibody::h3_filter            false
-antibody::h3_filter_tolerance  5

#more standard settings, for packages used by antibody_H3
-extrachi_cutoff 0

#these are standard settings for kic/ngk
-loops:legacy_kic false
-loops:kic_min_after_repack true
-loops:allow_omega_move true    ### remove 'true' and loop::?
-kic_bump_overlap_factor 0.36
-loops:refine_outer_cycles 5

#These enable the kink constraints.  Increase the weight if you want tighter kink constraints.
-constraints:cst_weight 1.0

The typical runtime varies, based on CDR H3 length (due to KIC). In our benchmark, the runtime was ~1 hour per model. We highly recommend using a cluster to speed up calculations.

Additionally, models should be validated for a reasonable VH–VL orientation. This can be done with following command: python $ROSETTA/main/source/scripts/python/public/plot_VL_VH_orientational_coordinates/

Deprecated Python-Based Grafting Approach

To build an antibody model from sequences of its light chain and heavy chain, you need

  1. your input Fv sequences
  2. (Downloading from developer-only repository: )
  3. ProFit (Installing ProFit3.1: )
  4. BLAST (C++ version) (Installing BLAST: )
  5. Rosetta

Currently, antibody homology modelling is a 3 step process:

  1. Template selections by BLAST
  2. Grafting of CDR templates onto a FR template and Fv refinement
  3. Intensive H3 modeling and VL/VH refinement

Steps 1 and 2:

This command should be all in one line. Separated for documentation

./ --light-chain <input_l.fasta> --heavy-chain <input_h.fasta> 
--superimpose_profit <ProFit path> 
--blast <blastp path> 
--blast-database <blast_database path (in tools/antibody)> 
--antibody-database <antibody_database path (in tools/antibody)> 
--rosetta-bin <rosetta/rosetta_sourse/bin> 
--rosetta-database <rosetta_database> 
--rosetta-platform <rosetta binary extension if not gcc. Example: linuxclangrelease or static.linuxclangrelease>

<input_l.fasta> and <input_h.fasta> are the files having sequences of the light and heavy chains, respectively, which you want to model.


  1. Sequence of the light chain Fv in FASTA format
  2. Sequence of the heavy chain Fv in FASTA format


  1. Sequence-grafted and refined Fv pdb: grafted.pdb, grafted.relaxed.pdb
  2. Constraints file for optional use in Step 3: cter_constraint

The script calls two Rosetta executable (relax and antibody_assemble_CDRs) for grafting and refinement, respectively, by specifying “–rosetta-bin” option. You can see other options by typing: ./ –help

Step 3:

[path to executable] /antibody_model_CDR_H3.[platform|linux/mac][compile|gcc/ ixx]release –database [path to database] @options

Sample options for a production run may look like: (this is an example, see details in RosettaAntibody3 application: Antibody Modeler Protocol (Loop H3 and VL-VH) .

Flags starting from "-kic_bump_overlap_factor 0.36" to "-loops:outer_cycles 5" will turn on the NGK or KIC2 for H3 loop modeling. Without these flags, the code is running KIC1 for H3

  -nstruct 2000   
  -s grafted.relaxed.pdb                             # Output of the
  -antibody::remodel              perturb_kic        # low-res H3 modeling
  -antibody::snugfit              true               # VL-VH orientation optimization via docking
  -antibody::refine               refine_kic         # high-res H3 modeling
  -antibody::cter_insert          false              # H3 cterminal insertion using Kink/Extend fragments
  -antibody::flank_residue_min    true               # minimize 2 stem residues each side of H3 during modeling
  -antibody::bad_nter             false              # if n-terminal stem of H3 is bad and you have a pdb file with correct stem to copy
  -antibody::h3_filter            true               # using bioinformatics rules of Kink/Extend to filter out bad H3 decoys
  -antibody::h3_filter_tolerance  20                 # the maximum number of filtering is set to 20
  -ex1 -ex2 -extrachi_cutoff 0                       # packing options
  -constraints:cst_file cter_constraint              # constraint file which can include one or two lines of below optional constraints: 
  -antibody:constrain_cter                           #   optional constraint (a) the H3 cterminus to be Kink/Extend
  -antibody:constrain_vlvh_qq                        #   optional constraint (b) the distance between two GLN-GLN residues one L and H chains
  -kic_bump_overlap_factor 0.36                      # KIC1 become KIC2 (or NGK) after turning on the flags from here
  -corrections:score:use_bicubic_interpolation false
  -loops:legacy_kic false
  -loops:kic_min_after_repack true
  -loops:allow_omega_move true
  -loops:outer_cycles 5


  1. Sequence-grafted and refined Fv pdb: grafted.relaxed.pdb
  2. Constraints file for optional use, output from steps 1 and 2: cter_constraint


  1. Set of modeled and refined Fv pdbs with loop modeled CDH3 loops: <grafted.relaxed_000X.pdb> We recommend generating at least 2000 decoys during this step

Post Processing

You can use a set of decoys simultaneously for antibody-antigen docking simulations, such as SnugDock and EnsembleDock.

See Also