This is SWISS-MODEL Beta server. Modelling hetero-oligomers is now possible!

You can input multiple target sequences for a single modelling job.
Enter the first sequence the normal way with copy and paste, or add a UniProtKB AC, then add another - and another - and another!
You can revert to the stable SWISS-MODEL website clicking here


Introduction to SWISS-MODEL

SWISS-MODEL is a web-based integrated service dedicated to protein structure homology modelling. It guides the user in building protein homology models at different levels of complexity.

Building a homology model comprises four main steps: (i) identification of structural template(s), (ii) alignment of target sequence and template structure(s), (iii) model building and (iv) model quality evaluation. These steps require specialized software and integrate up-to-date protein sequence and structure databases. Each of the above steps can be repeated interactively until a satisfying modelling result is achieved.

The SWISS-MODEL Workspace

The SWISS-MODEL Workspace (Biasini et al.) is a personal web-based working environment where several modelling projects can be carried out in parallel. Protein sequence and structure databases necessary for modelling are accessible from the workspace and are updated in regular intervals. Tools for template selection, model building, and structure quality evaluation can be invoked from within the workspace.

From the workspace the user accesses the current Modelling Projects, inspects their status and visualizes the results upon job completion. Project names can be changed retroactively by clicking on the symbol next to the project title. By default, projects are stored for two weeks on the server with an option to extend the project lifetime. The remaining time until a given project is deleted from the server is displayed accordingly.

Model Building

Models are computed by the SWISS-MODEL server homology modelling pipeline (Biasini et al.) and (Bordoli et al.) which relies on ProMod3, an inhouse comparative modelling engine based on OpenStructure (Biasini et al.).

Given a target-template alignment and a template structure, all conserved structural information is transferred to an initial model corresponding to the target sequence. Loop modelling is done using a database approach in order to find appropriate loop template candidates which are then adapted to the environment using CCD (Canutescu et al.). Side chains are modeled using a backbone dependent rotamer library (Shapovalov et al.) based on SCWRL4 (Krivov et al.) by the Dunbrack lab. A last energy minimization step is performed using the OpenMM (Eastman et al.) molecular mechanics library. The final model is selected based on model quality estimation employing statistical potentials of mean force.

Modelling Modes

Depending on the difficulty of the modelling task, three different types of modelling modes are provided, which differ in the amount of user intervention: automated mode, alignment mode and project mode.

Automated Mode

The Automated Mode only requires the amino acid sequence or the UniProtKB accession code of the target protein as input.

The automatic pipeline identifies suitable templates based on Blast (Altschul et al.) and HHblits (Remmert et al.). Template quality is then estimated from its properties (Biasini et al.). Priority is given to templates that maximize the expected quality of the models and the coverage of the target. In cases where templates mapping to the same sequence segment exhibit significant conformational differences, several models are generated to reflect the structural diversity.
Please note that templates for hetero-oligomer modelling are currently selected only based on the template quality (Biasini et al.). Coverage is not taken into account.
This mode is subject to continuous evaluation within the Continuous Automated Model Evaluation CAMEO (Haas et al.) platform.

Please note that it is unnecessary to run automated mode by pressing "Build Model" and afterwards start the project again and "Search for Templates" only. Both options start the same template search, which is also accessible in the first case once the models are built.

Alignment Mode

If the desired template for modelling is known and available in the SWISS-MODEL Template Library (SMTL), a target-template alignment in either FASTA or ClustalW format may be used to start the modelling process, thereby skipping the template search.

The template sequence(s) should be named using the PDB ID format (i.e., “1CNR” or “1CNR_A”). The user will be asked to specify which sequence in the alignment corresponds to the target and/or the template protein from a drop down list.

The Alignment mode allows the advanced user to invoke the modelling step starting from alternative alignments and evaluate the quality of these alternative models.


Project Mode

In difficult modelling situations, where the correct alignment between target and template cannot be clearly determined by sequence based methods, visual inspection and manual manipulation of the alignment can help improving the quality of the resulting model significantly.

The program DeepView - Swiss-PdbViewer (Guex et al., 1997) can be used to generate, display, analyse and manipulate modelling project files in the SWISS-MODEL workspace. Project files contain the superposed template structures and the alignment between the target and the template. In this mode the user has full control over essential modelling parameters, i.e. the choice of template structures, the correct alignment, and the placement of insertions and deletions in the context of the 3D structure. Project files can also be generated by the workspace template selection tools.

The program DeepView can be downloaded freely from the ExPASy web site. SWISS-MODEL supports DeepView legacy projects by relying on the previous version of the PROMOD modelling pipeline. It does not require administrator privileges for installation. Under MS Windows, simply uncompress the distributed archive at any location you like (e.g. c:\spdbv or on your desktop) and run the spdbv.exe application.

Ligand Modelling

Biologically relevant ligands and cofactors are modelled using a conservative homology transfer approach from the templates identified in the SMTL. Ligands in SMTL are annotated either as: (i) relevant, non-covalently bound ligand, (ii) covalent modifications or (iii) non-functional binders (e.g. buffer or solvent). A non-covalently bound ligand is considered for the model if the coordinating residues are conserved in the target-template alignment. The relative coordinates of the ligand(s) are transferred from the template, if the resulting atomic interactions in the model are within the expected range for van der Waals interactions and water mediated contacts.

Oligomeric Modelling

In SWISS-MODEL, the quaternary structure annotation of the template is used to model the target sequence in its oligomeric form. The method (Bertoni et al., in preparation) is based on a supervised machine learning algorithm, Support Vector Machines (SVM), which combines interface conservation, structural clustering and other template features to provide a quaternary structure quality estimate (QSQE). The QSQE score is a number between 0 and 1, reflecting the expected accuracy of the inter-chain contacts for a model built based a given alignment and template. Higher numbers indicate higher reliability. This complements the GMQE score which estimates the accuracy of the tertiary structure of the resulting model.

The SWISS-MODEL Template Library (SMTL)

The SWISS-MODEL template library is a large structural database of experimentally determined protein structures derived from the Protein Data Bank (Berman et al).

It serves as the main repository of structural information for the modelling pipeline and provides atomic coordinates of protein structures and maintains sequence and profile databases which can be searched by BLAST and HHblits. Alignment-independent properties of the template are precalculated and stored in the database, e.g. a mapping between residues resolved in the experiment and the corresponding residue in the full protein sequence, or predicted solvent accessibility and secondary structure information.

Individual entries of the SMTL can be inspected using the web interface. The sequence features are linked to a 3D structure viewer and can be interactively explored.

The web interface also includes an online annotation system for ligands contained in the experimental structures. Ligands can be marked as synthetic, natural or part of crystallization buffer. This information is used by the modelling pipeline to determine whether a ligand is considered for inclusion into the final model.


Biological Assemblies (Biounit) of Templates

The biological assembly, or biounit, describes the oligomeric state, or quaternary assembly, which is thought of as the biologically most relevant form of the molecule. For a detailed description see Biological Assemblies on PDB-101.

The biological assembly reported in the SMTL is retrieved from the PDB entry.

SMTL entries are organized (if more than one assembly is available) by likely quaternary structure assemblies which are created according to the author and software annotated oligomeric states listed in the PDB deposition. If not all chains of the asymmetric unit are included in the biounit, the asymmetric unit is included in the template.

Input Data

Protein amino acid sequence or UniProtKB identifier

The amino acid sequence of the target protein can be submitted either as a plain text, or in FASTA format.

- Example of plain text sequence:


- Example of FASTA sequence:

>sp|P00321|FLAV_MEGEL Flavodoxin - Megasphaera elsdenii.

If the protein sequence is deposited in the UniProtKB (Bairoch et al.) database, the UniProtKB identifier of the entry can be provided as input (i.e.: P00321). In this case, the identifier is immediately validated and replaced with the corresponding sequence.

Target-template alignment

The following formats are currently supported: FASTA and ClustalW.

- Examples for FASTA:


- Examples for Clustal:

CLUSTAL W (1.82) multiple sequence alignment
                .:***  ..*  :  **: * .. :**  :** **..: **  *   

User Template

Depending on the planned model application, it might be desirable to use a different structural template than the ones provided by the template search. In this case the user is required to provide a file in PDB format (*), making sure that it does not contain chemically modified amino acids. Oligomeric templates are accepted, with the requirement that homo-oligomer template chains must have the exact same amino acid sequence.

(*) A PDB-like file containing the coordinates of the template structure. For more information about PDB file format please see this link.

Please notice that the mmCIF format is currently not supported.

DeepView Project

Project files containing the superposed template structures, and the alignment between the target and template can be directly uploaded into the SWISS-MODEL Workspace. See the “Project Mode” section for further details.

An example of DeepView Project and its application in modelling of Oligomeric proteins can be found here.

Oligo Project (Beta)

The Oligo Project mode is dedicated to the modelling of protein quaternary structures, by exploiting the structural information available from homologous complexes.

The "Add Target" button is provided to input multiple protein chains or subunits. Target sequences can be submitted as a plain text, FASTA sequences, or UniProtKB AC's.

Immunoglobulin sequence input

If an antibody sequence is present in the input, the user is presented with a notice as well as a link to a dedicated server for antibody modelling (PIGSPro). By clicking on the link, the user is redirected to the PIGSPro server home page where the input form is pre-filled with the detected antibody variable domains. The link is displayed while the SWISS-MODEL pipeline continues running and thus the user has still the option to use SWISS-MODEL for the modelling.
An “Antibody detected” label is also shown on the different pages of the project.
Antibody sequences are identified by aligning the sequence against Hidden Markov Models developed on purpose for immunoglobulins.

Display of modelling results

Coordinates of the model, the corresponding alignment and quality evaluations can be accessed and downloaded via web-browser from the workspace.

Model details

This section allows to display the three-dimensional structure of model, the target-template sequence alignment and to download the model coordinates. For better assistance, many sequence features/scoring schemes are synchronised with the 3D molecular view.

The colouring of the alignment can be changed by clicking on the "Options" button (cog icon) and selecting the desired colouring scheme.

Model coordinates are available in two different formats:

If the model has been build using the Automated Mode, the information about the selected template(s) is provided with cross references to structural databases via the link to the SWISS MODEL Template library.

By default, the final model is presented coloured based on the QMEAN model quality. This allows instant visualisation of regions of the model that are well or poorly modelled. Information about the oligomeric state, as well as bound ligands and cofactors are provided. The user can alternatively choose to see the results in a well formatted report page which shows all the results in a readable format that can be copied and pasted to other documents. The user can download an archive file containing all the models and reports for the given target sequence.


Model evaluation


GMQE (Global Model Quality Estimation) is a quality estimation which combines properties from the target-template alignment. The resulting GMQE score is expressed as a number between 0 and 1, reflecting the expected accuracy of a model built with that alignment and template. Higher numbers indicate higher reliability. Once a model is built, the GMQE ((1) in the figure above) gets updated for this specific case by also taking into account the QMEAN score of the obtained model in order to increase reliability of the quality estimation.


QMEAN (Benkert et al.) is a composite scoring function based on different geometrical properties and provide both global (i.e.for the entire structure) and local (i.e. per residue) absolute quality estimates on the basis of one single model.

The QMEAN Z-score ((2) in the figure above) provides an estimate of the ‘degree of nativeness’ of the structural features observed in the model and indicates whether the model is of comparable quality to experimental structures. Higher QMEAN Z-scores indicates better agreement between the model structure and experimental structures of similar size. Scores of -4.0 or below are an indication of models with very low quality, this is also highlighted by a change of the "thumbs-up" symbol to a "thumbs-down" symbol next to the score.

QMEAN consists of four individual terms. The four individual terms of the global QMEAN quality scores are also listed ((3) in the figure above). The white area in the bar-plots (numerical values close to zero) indicates that the property is similar to what is observed in experimental structures. Positive values indicate that the model scores higher than experimental structures on average, negative numbers indicate that the model scores lower than experimental structures on average. The QMEAN Z-score itself is shown on top. The individual Z-scores compare the interaction potential between Cbeta atoms only, all atoms, the solvation potential and the torsion angle potential. For details please refer to the publication.

The “Local Quality” plot ((4) in the figure above) shows, for each residue of the model (reported on the x-axis), the expected similarity to the native structure (y-axis). Typically, residues showing a score below 0.6 are expected to be of low quality. Different model chains are shown in different colours. If the model is downloaded, the local score is reported in the B-factor column of the PDB file. The local quality can also be visualized by choosing the colour scheme "QMEAN".

In the “Comparison” plot ((5) in the figure above), model quality scores of individual models are expressed as 'Z-scores' in comparison to scores obtained for high-resolution crystal structures. The x-axis shows the length (in amino acids) of the proteins. The y-axis is the normalized QMEAN score. Every dot represents one protein structure. The darkest dots are all structures with a global QMEAN Z-score (the same score as 2 and 3 in the figure above) between -1 and 1, structures with a |Z-Score| between 1 and 2 are grey and if the |Z-Score| is more than 2 they are in light grey. A red star represents the model.

Modelling report

The SWISS-MODEL Homology Modelling Report offers a summary of all Models built in the project.

Note: The report is accessible (i) per model via a drop down menu, next to the model in the Models view or (ii) for all models in report.html in the downloaded file when choosing to download the project by pressing the download button below the project title.

It is structured in the following sections:

  • Model building Report: Contains project name, project date and references. The target sequence is in Table T1 of the Report.
  • Results: Version of the SWISS-MODEL template library and PDB release. All identified templates are listed in Table T2.
  • Models: Models are listed sequentially with each entry showing a picture of the model, a link to the PDB file, the version of the modelling engine, the oligomeric state, the ligands (if any), the global model quality estimate and the QMEAN score
    A graphical representation of the QMEAN score and it's four terms separately, the local quality estimate plot and the comparison with non-redundant set of PDB structures are also provided. For the template, a link to the template itself is provided together with the following information: the title of the structure, the target sequence coverage, the sequence identity to the target, the experimental method used to obtain the structure (and the resolution, if applicable), the oligomeric state, the ligands (if any), the sequence similarity to the target, the template search method used.
  • Save Project Locally: Allows to download the project as a zip file.
    The main folder contains the Model report (report.html), images folder (banner for the Report) and the model folder. Each model has its own subfolder.