The antibody homology modeling API creates homology models of antibodies using the Cyrus NextGen Antibody structure prediction method.

This method takes as input an antibody heavy and light chain and returns models of the antibody fV fragment. See below suggestions for modeling things other than antibody fVs; those workflows may involve other APIs.

1 Quickstart
- 1.1 Command Line Examples
- 1.2 Python examples
2 Inputs
3 Options
4 Output
- 4.1 Output File Interpretation
5 Notes
- 5.1 Modeling Other types of antibodies
  - 5.1.1 VHH or nanobody
  - 5.1.2 scFv
  - 5.1.3 Antibody full structures

Quickstart

Command Line Examples

Submit an antibody HM job to model the variable domains of an antibody:

cyrus engine submit antibody-hm \
  --heavy-chain DIQMTQTTSSLSASLGDRVTFSCSASQDISNYLNWYQQKPDGTIKLLIYYTSSLRSGVPSRFSGSGSGTDYSLTINNLEPEDIATYFCQQYSRLPFTFGSGTKLEIKRADAAPTVSIFPPSSEQLTSGGASVVCFLNNFYPKDINVKWKIDGSERQNGVLNSWTDQDSKDSTYSMSSTLTLTKDEYERHNSYTCEATHKTSTSPIVKSFNRNEC \
  --light-chain QVHLQESGPELVRPGASVKISCKTSGYVFSSSWMNWVKQRPGQGLKWIGRIYPGNGNTNYNEKFKGKATLTADKSSNTAYMQLSSLTSVDSAVYFCATSSAYWGQGTLLTVSAAKTTPPSVYPLAPGSAAQTNSMVTLGCLVKGYFPEPVTVTWNSGSLSSGVHTFPAVLQSDLYTLSSSVTVPSSPRPSETVTCNVAHPASSTKVDKKIVPR

Submit an antibody HM job to model an entire antibody:

cyrus engine submit antibody-hm \
  --model-full-antibody
  --heavy-chain DIQMTQTTSSLSASLGDRVTFSCSASQDISNYLNWYQQKPDGTIKLLIYYTSSLRSGVPSRFSGSGSGTDYSLTINNLEPEDIATYFCQQYSRLPFTFGSGTKLEIKRADAAPTVSIFPPSSEQLTSGGASVVCFLNNFYPKDINVKWKIDGSERQNGVLNSWTDQDSKDSTYSMSSTLTLTKDEYERHNSYTCEATHKTSTSPIVKSFNRNEC \
  --light-chain QVHLQESGPELVRPGASVKISCKTSGYVFSSSWMNWVKQRPGQGLKWIGRIYPGNGNTNYNEKFKGKATLTADKSSNTAYMQLSSLTSVDSAVYFCATSSAYWGQGTLLTVSAAKTTPPSVYPLAPGSAAQTNSMVTLGCLVKGYFPEPVTVTWNSGSLSSGVHTFPAVLQSDLYTLSSSVTVPSSPRPSETVTCNVAHPASSTKVDKKIVPR

Python examples

Submit an antibody HM job to model the variable domains of an antibody:

from engine.antibody_hm.client import AntibodyHmClient

client = AntibodyHmClient()
job_id = client.submit(
  heavy_chain="DIQMTQTTSSLSASLGDRVTFSCSASQDISNYLNWYQQKPDGTIKLLIYYTSSLRSGVPSRFSGSGSGTDYSLTINNLEPEDIATYFCQQYSRLPFTFGSGTKLEIKRADAAPTVSIFPPSSEQLTSGGASVVCFLNNFYPKDINVKWKIDGSERQNGVLNSWTDQDSKDSTYSMSSTLTLTKDEYERHNSYTCEATHKTSTSPIVKSFNRNEC",
  light_chain="QVHLQESGPELVRPGASVKISCKTSGYVFSSSWMNWVKQRPGQGLKWIGRIYPGNGNTNYNEKFKGKATLTADKSSNTAYMQLSSLTSVDSAVYFCATSSAYWGQGTLLTVSAAKTTPPSVYPLAPGSAAQTNSMVTLGCLVKGYFPEPVTVTWNSGSLSSGVHTFPAVLQSDLYTLSSSVTVPSSPRPSETVTCNVAHPASSTKVDKKIVPR")

Submit an antibody HM job to model an entire antibody:

Inputs

Heavy and light chain sequences to model (canonical amino acids only) to model

Options

--heavy-chain
- a protein sequence
- Only canonical amino acids are supported.
- Please include only the CH domain sequence. Further heavy chain sequence will be ignored.
--light-chain
- a protein sequence
- Only canonical amino acids are supported.
- Please include only the CL domain sequence. Further light chain sequence will be ignored.
--model-full-antibody
- boolean
- Default is false, if this flag is set to true the entire antibody will be modeled. If false, only the variable domains will be modified

Output

models (directory)
- 5 PDB files representing the centers of the top-scoring clusters of models generated during the homology modeling process.

Output File Interpretation

Cyrus’s Antibody HM tool returns 5 cluster centers (selected_models.tar.gz) after running a large number of HM trajectories. This clustering is balanced to return 5 models that have good energy within their structural cluster and represent different clusters.

If all 5 models are similar even after clustering, it means that HM was highly converged and/or that the template match was very high. This is a good sign, it means Rosetta has good confidence in this prediction.

If there are 5 distinct predictions, particularly models that vary outside the H3 loop, it may mean that the default sampling is insufficient, or that this particular problem is harder than this API is able to accommodate – please let Cyrus know and we can discuss other options for this type of modeling problem.

If your models are highly similar except for the H3 loop, consider using your preferred structure as input to loop modeling for further refinement.

Depending on settings, you may also see full_selected_models.tar.gz. If you did not use --model-full-antibody please ignore this.

Notes

Modeling Other types of antibodies

VHH or nanobody

VHH or nanobody antibodies are composed of single chains and thus compatible with the single-chain HM API service. The antibody specializations in this tool are focussed on the VH/VL orientation and thus irrelevant to VHH antibodies. Further loop refinement of CDRs can be performed with the loop remodel tool.

scFv

Often, the linker for an scFv is not an important part of the problem. Many scFvs have off-the-shelf, stock linkers designed to not interfere with the underlying Fv. If you wish to explicitly model the linker, use this tool to model the Fv, and use the resulting model as a custom template in the single-chain HM API service to fill in the linker. Further loop refinement of the linker can be performed with the loop remodel tool.

Antibody full structures

--model-full-antibody can be used as a flag to this API to return full-length antibodies instead of just the scFv. The remaining chains are modeled with a multichain homology modeling tool.