Kotai Antibody Builder is a state-of-the-art antibody Fv modeling server established through a collaboration between Astellas Parma Inc., Osaka University, and the National Institute of Biomedical Innovation (NIBIO). Kotai Antibody Builder constructs three-dimensional (3D) structures of antibody variable domains from sequence using canonical rules, new H3-rules and evolutionary information. This web service is a fully automated version of the semi-automated pipeline used successfully in the Second Antibody Modeling Assessment (AMA-II).
Individual sequences, either plain text or FASTA format, of heavy and light chains for antibody Fv region are accepted as input.
When you submit a job, a link will appear where the output will be shown. After the job has finished, you can visit the link to view the results, or download the model, alignment, template and log files.
You can find a detailed description of our original pipeline in the AMA-II paper. The implementation in the web server is somewhat different. Here we describe a brief introduction of our fully automated protocol.
Astellas Parma Inc. has developed a module called MANGO for this purpose. The framework templates are selected by sequence identity of the whole molecule, RamaFavored score (>85%) from Molprobity, number of in/dels in the framework region, resolution of crystal (< 2.8A), and framework canonical classification of heavy chain (that is, the classification of backbone structure based on the sequence). We utilized the definition of canonical structures and key residues proposed by Dunbrack et al.  to select the non-H3 CDR templates. The canonical templates are sorted by the sequence identity of the loops, resolution, R-free and B-factor. In addition, PSSM-based scoring to predict canonical structures is also used. If both methods predict different classifications, the PSSM prediction is preferred. When the framework has non-H3 CDRs in the same canonical classification as that predicted for the query, the server will not graft the loop in the framework. Although the CDR-H3 doesnt have a canonical structure, our group has been developing H3-rules [4-6], which can predict the form of the base region of the loop from the pattern of amino acids. If the loop length is more than 5, the server searches for appropriate templates using key residues in the query sequence. After filtering possible templates in terms of the loop length and H3-rules, the sequence identity of the loop, resolution, R-free and B-factor are considered further.
By default, the server grafts CDR templates to the framework template, and then runs energy minimization using the cosgene module in myPresto suite ver. 4.0 . The AMBER96 forcefield  is used with positional restraints to the backbone atoms followed by OSCAR-star side-chain modeling . Further sampling of CDR-H3 conformations by fragment assembly is available as an option with the number of models to be generated chosen by the user. A refinement process is then carried out comprised of side-chain modeling by OSCAR-star and successive energy minimizations by cosgene and OSCAR-loop . The resulting OSCAR-loop scores, CDR-H3 kinked or extended classifications of the resulting loops will be displayed. Alternative sampling by multi-canonical molecular dynamics (McMD), as used in AMA-II, will be available in the future.