SAbPred Help


Hybrid Loop Structure Prediction


Input

To model a loop sequence:
  1. Go to the General Protein Sphinx or H3-specific Sphinx.
  2. Upload a structure file WITHOUT the loop you wish to model.
  3. Input the chain on which the loop to model is supposed to be.
  4. Input the the number of the residue BEFORE the beginning of the loop. This residue is always present in the input model.
  5. Input the number of the LAST residue of the loop. This does not have to be present in the input model.
  6. Click "Submit"

Process

  • The Sphinx algorithm will produce decoys for the loop using a hybrid of database search and ab initio modeling.
  • The top 500 decoys will be selected by scoring via the SOAP Loop (Dong et al., 2013).

Output

  • Visualization of the top 500 loops.
  • all_decoy_structures.pdb - literally, all the decoy structures that were generated by Sphinx during the modelling process.
  • logfile.txt - a file containing the job information and details of each fragment used (alignment, ESS score etc.).
  • ranking_results.csv - the scores for all the decoys generated, calculated using Sphinx's inbuilt ranking function.
  • top_500_decoy_structures.pdb - the structures of the top 500 decoys, as ranked using Sphinx's inbuilt ranking function.
  • CompleteModels - an archive containing the full PDB files for the top 500 decoys.
  • SOAP-Loop_results.csv - a ranking of the top 500 decoys, carried out using the SOAP-Loop potential of MODELLER.

Antigen receptor Numbering and Receptor ClassIfication is a tool to annotate antibody and T-cell receptor sequences.


Input

To number an antibody or TCR variable domain sequence:
  1. Go to the Sequence Numbering page.
  2. Paste a single amino-acid sequence into the text box or upload multiple sequences in a fasta file.
  3. Click "Annotate"
  4. [Optional] Select the numbering scheme used to annotate sequences with.
  5. [Optional] Select output format of the annotation (plain txt file or a excel compatible csv file).
  6. [Optional] Restrict recognised domains to antibody domains. Unchecking this box with enforce the imgt numbering scheme to be used

Process

  • The ANARCI algorithm will align each sequence to a database of Hidden Markov Models that describe the germline sequences of domain types from different species.
  • HMMER3 is used to perform the alignment
  • Where a domain is recognised, the most significant alignment is used to classify it.
  • The alignment to the HMM's reference states is then used to annotate the sequence with the desired numbering scheme.

Output

  • Each recognised variable domain in the submitted sequence will be listed along with its species, domain type, starting position in the sequence (0 is the first position) and ending position.
  • The numbering file in the format you selected is available for download
  • When multiple sequences are uploaded, the recognised domains for each sequence is shown

To see a demonstration of using the modelling server, check out our video!

Input

To create a model of your antibody:

  1. Go to the Antibody Modelling page.
  2. Submit the amino-acid sequences of your desired antibody (heavy and light chains). This can be done by copying and pasting the amino-acid sequence for each chain into the corresponding text-box.
  3. Name your model in the "Job Name" box (only alphanumeric characters, spaces and the characters +_.- can be used).
  4. [Optional] Choose which numbering scheme should be used to annotate the final model structure(s). The Kabat, Chothia, Martin (Enhanced Chothia) and IMGT numbering schemes are available.
  5. Click on the "Model" button to start.

Pipeline

The modelling process is fully automated. When you click "Model", the following steps will be performed:

  • The heavy and light sequences are initially numbered using the IMGT numbering scheme via ANARCI
  • The North/AHo definition of CDR loops and framework regions is used throughout.
  • Templates for the VH and VL domain framework are chosen from SAbDab based on sequence identity.
  • The VH-VL orientation is chosen using ABangle and the VH and VL templates are orientated in the chosen pose
  • CDR loops are modelled using CDR-specific FREAD. If no acceptable template is found, then a second round of searching is performed on antibody-specific FREAD. If this still does not yield a suitable template, this is noted and a separate template is chosen based on sequence similarity and length. If necessary, MODELLER is used to model loops ab-initio.
  • Once assembled, PEARS is used to predict the side chains.
  • The final model is analysed and renumbered with the chosen scheme. A modelling report is generated that estimates how accurate each region is given the choices made, methods used and scores calculated when building your model.
On submission, you will be taken to a modelling results page. From here you can check the status of your job and view snapshots of the log file. Multiple jobs can be monitored from the Jobs page.

Output

An example output can be seen here. Output for Modelling consists of:

  • Summary Information
    • The log file detailing the decisions made through the modelling process
    • A sequence annotation file giving the numbering of your submitted sequence
    • An archive zip file containing all the models, templates and modelling details for the job.
  • Models. The model ranked "1" is considered the best.
    • The model file contains the coordinates in PDB format.
    • The View Model Annotations link takes you to the model annotation page (see below).
  • Template details
    • A table listing the template region, the template structure, the selection method and the score. For FREAD results, the score is the environment specific substitution score. Otherwise the score is the sequence identity over the region.
    • An alignment between your submitted target sequence and the templates chosen. Use the selection at the bottom to change the definition of CDR (highlighted in red).
The results page allows you to annotate your model with structure and sequence based properties or to use it in SAbPred's paratope and epitope prediction applications. To do so click on the corresponding option in the 'Action' drop-down menu next to the model details.

Model Annotations

Once an Fv model has been predicted it is annotated with a range of properties [Example].

  • All numbering shown will be in the scheme you chose at the input page (Chothia, Kabat, IMGT [used in ABodyBuilder] or Martin/Enhanced Chothia).
  • A bio-PV visualisation of your model will be shown (we strongly recommend using Google Chrome to load this page). Click the buttons on the right hand side to change which type of annotation is shown, and the buttons on the left to change how the structure is represented. The available annotations are as follows:
    • Secondary structure. The molecule is annotated as either alpha helix, beta strand, beta turn or loop.
    • Domain and CDR definition. VH and VL domains are coloured green and cyan respectively. CDRs are coloured red. The CDR definition used can be either Kabat, Chothia, IMGT, Aho/North (this is used in ABodyBuilder) or Contact.
    • Solvent exposure. DSSP is used to calculate the accessible surface area. This value is scaled by the standard accessibilities to give the relative exposure. Residues that are less than 10% exposed are considered buried.
    • Estimated accuracy. Each of the choices that can be made in the modelling procedure has been benchmarked. We use this information to estimate how well each region of the model is likely to have been predicted. For example, 75% of VH framework structures with sequence identity of 80% +/- 2.5% have a backbone RMSD of 1Å or better. Therefore if the model has been predicted using a template with 80% sequence identity to the target we have a confidence of 75% that the VH framework is modelled to within 1Å RMSD. Use the sliders to change the thresholds.
    • Sequence liabilities. Sequence motifs that are known to cause issues for antibody development are mapped onto the structure. Use the buttons to toggle them on and off. Click the checkbox to hide those that are not exposed to the solvent.


For a full description of the method or if you use this software, please refer to:
Leem et al. ABodyBuilder: automated antibody structure prediction with data-driven accuracy estimation. mAbs (2016)

Input

To predict the side chains for your antibody, or mutate it:

  1. Go to the PEARS page.
  2. Upload the structure for side chain predction. You also have to specify which chain(s) form the antibody.
  3. [Optional] If you would like to mutate the antibody structure, then submit the sequence of the antibody. Use UPPER CASE letters for spots you would like to mutate and lower case elsewhere.
  4. Click on the "Model" button to start.

Algorithm

PEARS uses the IMGT position-dependent distribution of amino acid rotamers to predict side chains.

  • The heavy and light chains are initially numbered using the IMGT numbering scheme via ANARCI
  • PEARS initially builds the disulphide bridges of the antibody chains.
  • PEARS then searches for position-amino acid combinations (i.e. side chain types) that have a unimodal χ1 angle. At specific locations, some amino acids tend to have one χ1 angle.
  • Once these unimodal side chain types are predicted, PEARS predicts the remaining side chains using dead-end elimination and graph decomposition.
PEARS typically requires less than 10 seconds to complete the side chain prediction process.

Output

An example output can be seen here. Here the structure with the predicted side chains is you can download:

  • IMGT-renumbered structure. The input structure in the IMGT numbering.
  • PDB-numbered structure. The input structure in its original numbering, but with the new side chains.
  • Mapping files. Text files that correlate residue numbers from the input PDB file into the IMGT numbering.
CDR region (Red)
Light Chain (Green)
Heavy Chain (Yellow)
Antibody i-Patch predictions
Warmer color means higher likelihood to be a contact with the antigen

Paratope predictions are performed by Antibody i-Patch (AB i-Patch), a method to annotate the likelihood of antibody residues to be in contact with a given antigen. As input you will need a structure of the antibody (can be a model) and the antigen. Follow these steps to obtain your prediction:

  1. Go to the Paratope Prediction Submission page.
  2. Submit structures of the antibody and the antigen. For instance, you can use this structure which contains both antibody and antigen (submit it in both antibody and antigen fields). Alternatively supply a four digit PDB code to use a known structure.
  3. Specify the chains to be used for the antibody and the antigen (using the example of 1A2Y above, chains AB for antibody and C for antigen).
  4. Press submit, you should be taken to a 'waiting' page if there were no issues with yout input.
  5. The Waiting page will refresh every 10s, until you will see a finished (example) job.
At any time you can get see the Paratope Prediction jobs that are currently running in the jobs tab on the submission page. You can also see all the jobs running on the service here.

Output

An example output can be seen here. The output file for paratope prediction consists of:
  • The structure you have submitted with the B-Factor field replaced with the AB i-Patch score. The higher the i-Patch score the higher the confidence that the residue will be part of the paratope.
  • A visualisation of the structure with residues coloured by i-Patch score (PV or as a Pymol session file). Warmer colours correspond to a higher i-Patch score.
  • A visualisation of the sequence of the submitted antibody coloured by i-Patch score. You may also highlight the CDR regions according to different definitions. The details of the N residues with the highest i-Patch score can be exported by clicking on the 'Export top N' button. Choose a numbering scheme from the drop-down menu with which to annotate the exported residues.


For a full description of the method or if you use this software, please refer to:
Krawczyk et al. Antibody i-Patch prediction of the antibody binding site improves rigid local antibody‚Äďantigen docking. PEDS (2013)

EpiPred predicts structural epitopes specific to a given antibody. As input you will need a structure of the antibody (can be a model) and the antigen. Follow these steps to obtain your prediction:

  1. Go to the Epitope Prediction Submission page.
  2. Submit structures of the antibody and the antigen. For instance, you can use this structure which contains both antibody and antigen (submit it in both antibody and antigen fields). Alternatively supply a four digit PDB code to use a known structure.
  3. Specify the chains to be used for the antibody and the antigen (using the example of 1A2Y above, chains AB for antibody and C for antigen).
  4. Press submit, you should be taken to a 'waiting' page if there were no issues with yout input.
  5. The Waiting page will refresh every 10s, until you will see a finished (example) job.
At any time you can get see the Epitope Prediction jobs that are currently running in the jobs tab on the submission page. You can also see all the jobs running on the service here.

Output

An example output can be seen here. Output for epitope prediction consists of:
  • A structure and summary file for the top-ranking predictions (0 being the best)
  • B-Factor annotated structure. This is mainly for visualization purposes as the chains have been renamed and residues renumbered. You should be able to visualize the prediction if you color this structure by B-Factor in Pymol or similar. The predicted residues have B-Factor of 100 and all the others 0.
  • Summary file. This is the annotation of the residues predicted to be in the epitope, using numbering and chain annotation identical to your submitted file. Each line corresponds to a sequence id and the chain, thus a line '10 B' means that residue 10 on chain B has been predicted to be part of an epitope.
For a full description of the method or if you use this software, please refer to:
Krawczyk K.et al. Improving B-cell epitope prediction and its application to global antibody-antigen docking. Bioinformatics. (2014)
  • SAbPred is built and maintained by the Oxford Protein Informatics Group (OPIG)
  • Please see the FAQ and topics above for help using SAbDab
  • For any further queries please use our query form or contact us directly: opig <~at~> stats.ox.ac.uk