Why was SWISS-MODEL created in the first place?

Computational structural biology has made tremendous progress over recent decades, with many computational methods becoming a key contributor to modern life science research. At the time when SWISS-MODEL was created, 25 years ago, experimental structures were known for only around 1,000 different proteins (100 times fewer than today), and protein modelling was the domain of very few experts. It required specialized hardware and software and in-depth knowledge of modelling techniques.

swiss model logo

A major driver behind the development of SWISS-MODEL has been to mask this complexity behind a simple interface, making protein modelling accessible to the broader life science community. To this end, Manuel Peitsch wrote the first version of SWISS-MODEL, which made it possible to generate a 3D model of a protein simply by sending a protein's sequence by e-mail. Swiss-PdbViewer (aka DeepView) developed by Nicolas Guex, provided a user-friendly, interactive protein visualization software, and later shared the software backend with SWISS-MODEL. Torsten Schwede joined the team of Manuel and Nicolas at GlaxoWellcome experimental research in Geneva to further develop SWISS-MODEL as a fully automated modelling service for the biomedical research community.

As the technology evolved over time, the philosophy remained the same. Today, the current version of SWISS-MODEL uses the latest web-based techniques to provide intuitive and interactive access to modelling, results visualization and interpretation directly in a web-based workspace accessible via a regular web browser. The complexity of the modelling engine in the backend – consisting of an intricate network of software and databases running on powerful IT hardware – remains invisible to the user.

Today, SWISS-MODEL processes over one million model requests per year and is one of the most widely used structure modelling servers worldwide.

In a nutshell, SWISS-MODEL...
  • ...was created to make protein modelling accessible to non-experts, thanks to an intuitive and interactive interface;
  • ...was the first automated homology modelling service on the internet;
  • ...facilitates the understanding of molecular mechanisms by site-directed mutagenesis; the interpretation of disease-related mutations; the design of new drugs or the determination of experimental protein structure;
  • ... contains an up-to-date repository of high quality 3D models of the full proteomes of 12 model organisms, including humans and several pathogens;
  • ...is tightly interconnected with several other bioinformatics resources, at the national and international level.

What is 3D protein modelling used for? Could you explain how SWISS-MODEL contributes in that context?

Knowledge of the three-dimensional (3D) structure of a protein provides invaluable insights into the molecular basis of its function. This information is useful for a wide range of applications, such as the design of specific experiments aimed at understanding molecular mechanisms by site-directed mutagenesis, mapping disease-related mutations, or rationally designing inhibitors and drugs. While the usefulness of 3D structures is obvious, experimental structure determination is inherently time consuming, expensive and not always successful, with the result that for a large number of proteins, no experimental structures are available.

Computational methods have established themselves as a valuable tool, complementing experimental techniques to gain structural insights. Currently, the most accurate and widely used approach is known as “comparative” or “homology” modelling, where experimentally determined structures of evolutionary related proteins are used as templates to build three-dimensional models for proteins of interest (targets). SWISS-MODEL was a pioneer in the field of fully automated comparative modelling as a server, and has been continuously developed over time. The latest efforts have been devoted to extending the scope of comparative modelling to homo- and heteromeric complexes. One important aspect in this context is reliable model quality estimates – which are essential to decide if a predicted model is expected to be suitable for the intended application.

The optimized SWISS-MODEL engine makes it possible to generate a three-dimensional model of a protein in a few minutes of CPU time, which is particularly relevant for high-throughput, proteome-wide applications. Thanks to this efficiency, we are able to keep an up-to-date repository of high quality 3D models of the full proteomes of 12 model organisms, including humans and several pathogens.

Could you give us an example of a discovery made possible thanks to SWISS-MODEL and its biological significance?

With thousands of users annually, leading to around 2,000 citations each year, it is hard to pinpoint one representative example. The majority of SWISS-MODEL users come from the biology and biomedical fields and use models in various different ways as substitutes for experimental structures. Let’s try to showcase a few examples illustrating the wide range of research fields, where SWISS-MODEL is applied.

One typical application of models is to analyse the effects of sequence variations within a family of organisms. As a concrete example, researchers have used SWISS-MODEL to understand the evolutionary change in haemoglobin that allowed Andean house wrens to adapt to high altitudes. In this context, the interpretation of disease-related mutations is especially pertinent – a challenging task that is becoming even more relevant as genome sequencing becomes routine practice in health care.

Models are also frequently used to aid the experimental structure determination process itself, e.g. by providing high-quality models to fit into low resolution EM densities, or to serve as search models in molecular replacement to solve the phase problem of X-ray crystallography. Last but not least, thanks to its ease of use, robustness and reliability, SWISS-MODEL is frequently used in teaching and outreach activities all around the world, as exemplified by various events organized by SIB and ELIXIR, which are impacting the next generation of scientists.

SWISS-MODEL relies on other bioinformatics resources to operate, such as the protein function knowledgebase UniProtKB/Swiss-Prot – another SIB Resource. Apart from one-way technical cooperation (e.g. retrieval of annotations), are there other collaborative relationships in place?

Indeed, the bioinformatics community is a tightly interconnected ecosystem. SWISS-MODEL would not be possible without relying on several other bioinformatics resources – data as well as software tools – such as the worldwide protein data bank (PDB), UniProtKB/Swiss-Prot developed by SIB, or the HMM search tools developed by the Söding lab. In return, SWISS-MODEL enriches sequence-based annotations in UniProtKB and other resources with 3D model information for proteins where no experimentally determined structure is available. For example, this enables users of UniProtKB not only to get a 3D view of their protein of interest, but also to directly map residue level annotation such as disease-related mutations into a structural context. To give an idea, the SWISS-MODEL Repository currently contains close to 1.5 million high-quality models for UniProtKB sequences, compared with 138,072 structures from PDB, which are mapped to UniProtKB.

Similar exchanges take place with InterPro, NextProt and STRING, from where we additionally retrieve and display interaction network data for proteins. As part of a recently launched ELIXIR project, we are collaborating with the CATH group to improve the quality of models by including family-specific information and developing APIs for automated access to model information.

One frequently asked question is whether SWISS-MODEL can be used for structure-based drug design. In order to provide this functionality, we have been collaborating with the SIB Molecular Modelling Group to implement a direct link between SWISS-MODEL and the SwissDock service. This enables interested users to predict molecular interactions between a small molecule and a modelled protein.

Several of the tools developed in our group related to SWISS-MODEL are of general use for other groups, too. For example, CAMEO, a fully automated evaluation service for modelling servers, helps methods developers to benchmark their new methods; ModelArchive is a long term archive for theoretical models published in the scientific literature; and OpenStructure – a modular software framework for computational structural biology which forms the basis for all new developments in SWISS-MODEL. We are making these tools freely available to the community.

Looking ahead, what are the top three goals for the development of SWISS-MODEL?

SWISS-MODEL was the first automated homology modelling service on the internet, and today it is one of the most popular and widely used structure modelling services worldwide. However, the field is highly competitive, with hundreds of other groups developing similar methods. In order to stay competitive, we need to continuously improve our service. Our top three goals going forward are to advance the state of the art with respect to:

  • Providing highly accurate 3D homology models based on the latest available information and state-of-the-art algorithms, accompanied by reliable and robust model quality estimates;
  • Improving the biological relevance of protein models by predicting quaternary structures, essential ligands and cofactors, and relevant conformational states;
  • Extending the scope of SWISS-MODEL to new biomedical applications, while keeping the overall philosophy of an intuitive, reliable and robust user interface.

On 18 October, SWISS-MODEL held its very own symposium in Basel to celebrate its 25th birthday: what was the main outcome?

The symposium was a great opportunity to bring together world-class experts in computational structural biology and the next generation of young scientists. Obviously there was some aspect of “looking back” to the last 25 years of SWISS-MODEL, how it all started and what made it successful over such a long time. But most presentations explored the state of the art of different aspects of computational structural biology, such as understanding the function of enzymes, prediction techniques for protein complexes, GPCRs and antibodies, critical community benchmarking efforts, classifying proteins into functional families, interpretation of missense variants, and on predicting disorder in proteins.

With the historical perspective in mind, we discussed what relevant problems in the field are still open and unresolved. As such, the main conclusion is probably that computational structural biology is still a very active field of research with many exciting developments happening and new players (such as Google DeepMind) entering the field.

We are not even close to running out of relevant problems to solve over the years ahead.