The AMPS-NMR (AMBER-based Portal Server for NMR structures) (Bertini et al., 2011) provides a user-friendly interface to run restrained molecular dynamics simulations (rMD) to refine experimental NMR structures, and to perform molecular dynamics simulations of biomacromolecular systems in general. Calculations can be run on CPUs or GPGPUs. The latter provide significantly faster performance (Andreetto et al., 2017) but are limited in availability. For NMR structures, the use of the AMPS-NMR portal results in a consistent improvement of features such as rotamer distributions, backbone normality and occurrence of steric clashes, without affecting agreement with the experimental data. The effect of rMD refinement is especially relevant for protein regions where the amount of experimental information is scarce. The rMD protocol has been implemented as a predefined multi-step procedure, so that the user does not need to know in detail how to tune the many parameters involved in a relatively complex MD calculation; the portal also handles automatically the most commonly used formats for experimental data. The AMPS-NMR webserver is available at http://py-enmr.cerm.unifi.it/access/index.

The interaction between molecules can be experimentally characterized with methods such as cross-linking mass spectrometry (XL-MS) and FRET (Okamoto and Sako, 2017; Yu and Huang, 2017) which can provide information about which residues are participating in the interaction as well as the distance (or an upper limit to it) between the reacting groups of the residues. DISVIS is a tool that allows researchers to visualize the three-dimensional distance-restrained possible interaction space between the interacting proteins partners. It can be used to investigate if the restraints are mutually exclusive and identify possible false positives in the input data. DISVIS is freely available at https://wenmr.science.uu.nl/disvis.

FANTEN (Rinaldelli et al., 2015) aims at the determination of the anisotropy tensors associated with NMR paramagnetic pseudocontact shifts (pcs) and/or residual dipolar couplings (rdc’s). FANTEN also permits the pcs-driven determination of the 3D structure of protein-protein adducts using a rigid body approach; the user can then submit these models directly to HADDOCK for flexible refinement. The FANTEN web server is available at http://abs.cerm.unifi.it:8080/.

The pioneer of integrative docking, HADDOCK (High Ambiguity Driven biomolecular DOCKing) (Dominguez et al., 2003) allows users to make use of a variety of experimental and theoretical information to drive the docking of up to 20 macromolecules. The latest version (v2.4) supports the docking of proteins, small-molecules, nucleic acids, peptides, cyclic peptides, glycans, and glycosylated proteins. To accommodate the complexity that naturally arises from >18 years of development of new features and docking modes without compromising usability and accessibility, HADDOCK has been made freely available (with registration) to the community since 2008 via an intuitive webserver (Vries et al., 2010; Zundert et al., 2016). Since exposing 400+ parameters could hinder user experience (and usability), users are separated in access tiers: Users belonging to the “Easy” tier can access and modify a limited number of parameters whereas users belonging to the “Guru” tier (granted upon specific request from users in their registration page) have full access and control of the docking protocol. The HADDOCK webserver is available at https://wenmr.science.uu.nl/haddock2.4.

MetalPDB (Andreini et al., 2013a; Putignano et al., 2017) is a database collecting information on the metal-binding sites present in the 3D structures of biological macromolecules (Putignano et al., 2017). In particular, metal sites are represented in MetalPDB, as 3D templates describing the local environment around the metal ion(s) independently of the overall macromolecular fold. Among its features, MetalPDB provides detailed statistical analyses regarding metal usage in proteins of known 3D structure, encompassing protein families, and in catalysis; it also includes tools for the structural comparison of metal sites (Andreini et al., 2013b). MetalPDB is available at https://metalpdb.cerm.unifi.it/

The classic Protein Data Bank (PDB) file format is a flat text file that is still used by many structural biology software to represent the spatial coordinates of macromolecular structures, even though the official format of the worldwide PDB (wwPDB) (Berman et al., 2003) has now shifted to the Crystallographic Information Framework (mmCIF) dictionary (Bourne et al., 1997). The simple flat text PDB format, despite its apparent simplicity, follows strict formatting rules which can be easily overlooked during editing resulting in various errors. Such errors are usually of syntactic nature (e.g., shifting columns or missing fields) due to typos introduced by users after manual-editing PDB files, although semantic errors are also very common (e.g., incorrect atom naming, overlapping numbering, duplications, etc.). PDB-tools web (Jiménez-García et al., 2021) is a fully configurable, user-friendly web interface for the command-line application pdb-tools (Rodrigues et al., 2018). Using the portal users can, in a few clicks, build a complex pipeline which can then be saved (and uploaded) for future use and reproducibility. This pipeline is composed by different processing blocks with atomic tasks on the PDB provided data, to name a few, removing certain atom types, renaming chains, renumbering residues, removing multiple occupancies or detecting gaps among many others. The webserver is available at https://wenmr.science.uu.nl/pdbtools/ and the list of all available processing blocks is described on the online manual at https://wenmr.science.uu.nl/pdbtools/manual.

Cryo-electron microscopy (cryo-EM) is an experimental technique that can be used to obtain direct images of large macromolecular complexes (Fernandez-Leiro and Scheres, 2016). When the resolution of the resulting EM data is not sufficient to build models directly in the maps, it is common to fit existing 3D structures into those maps. To this end, Powerfit has been developed as a Python package for fast and sensitive rigid body fitting of molecular structures into EM density maps with a core-weighted local cross correlation scoring function (Zundert and Bonvin, 2015b). It is freely available as a webserver (Zundert et al., 2017) running on both local and GPU-accelerated distributed infrastructures at https://alcazar.science.uu.nl/services/POWERFIT.

Monoclonal antibodies have been consolidated as therapeutic tools due to their high affinity and specificity towards their target. proABC-2 (Ambrosetti et al., 2020), the updated version of proABC (Olimpieri et al., 2013), is a valuable tool that predicts which antibody residues can form intermolecular contacts with its antigen and further give insight into the chemical components of such interactions, differentiating between hydrophobic and hydrophilic interactions. The latest version replaced the previously used random forest paratope predictor by a convolutional neural network algorithm. Predictions obtained by proABC-2 can be used for example, to guide molecular docking, leading to improvements of success rate and quality of the predicted models. proABC-2 is available at https://wenmr.science.uu.nl/proabc2/.

In addition to elucidating the structural aspects of a biomolecular complex and its interacting components, it is of crucial importance to be able to estimate the strength of the interaction, or, in other words, to be able to predict the binding affinity of the complex. PRODIGY, the PROtein binDIng enerGY Prediction tool (Vangone and Bonvin, 2015) allows to predict the binding affinity of protein-protein complexes from their three-dimensional structures. It does so by calculating features derived both from interfacial residue contacts as well as those from the non-interface residues (Kastritis et al., 2014). The same rationale has been expanded to protein-ligand complexes (PRODIGY-LIGAND) (Kurkcuoglu et al., 2018; Vangone et al., 2018) and to the classification of interfaces in crystal structures as biological or crystallographic (PRODIGY-CRYSTAL) (Elez et al., 2018; Jiménez-García et al., 2019). PRODIGY services are available at https://wenmr.science.uu.nl/prodigy/.

Once an interaction has been characterized, it is valuable to know which parts of the interface contribute more to the binding, namely what are interaction “hot spots”—residues that upon mutation to alanine confer a difference in the binding free energy of the complex greater than 2.0 kcal/mol. The experimental identification of these regions is typically low throughput, laborious and expensive since it involves multiple rounds of point mutations and binding energy evaluations. SpotON was developed a computational approach to identify and classify interface residues as either hot spots, with positive contribution to the interface or null spots, with little to no significant contribution to the interface (Melo et al., 2016; Moreira et al., 2017). Its predictor combines different machine-learning algorithms, and it is available as a free and easy-to-use webserver https://alcazar.science.uu.nl/services/SPOTON.

The EGI Federation is an international e-Infrastructure providing advanced computing and data analytics for research and innovation. In the last decade it evolved from the high energy physics compute grid (WLCG) towards a multi-disciplinary, multi-technology infrastructure federating hundreds of resource centers worldwide and delivering large-scale data analysis capabilities (>1 millions of CPU-cores and >900 PB of disk and tape storage) to more than 70,000 researchers covering many scientific disciplines. During 2020 it delivered 13 billions of CPU hours of High Throughput Computing (HTC) and 18 millions of CPU hours of cloud computing (Figure 2).

EGI was one of the main actors and the coordinator of EOSC-Hub, a three-year European project that ended in March 2021, aimed at starting the design and implementation of the European Open Science Cloud. EGI operates several core elements of EOSC and contributes to EOSC service portfolio offering services like e.g., HTC, Cloud Compute, Workload Manager, and Check-in which are relevant for WeNMR. Moreover, with the support of the recently started EGI-ACE project, under which WeNMR services are further operated, EGI will deliver the EOSC Compute Platform and will contribute to the EOSC Data Commons through a federation of cloud compute and storage facilities, Platform as a Service (PaaS) services and data spaces with analytics tools and federated access services.

The collaboration between EGI and WeNMR dates back to their early days, when in 2011 WeNMR became the first Virtual Research Community officially recognized by EGI. A formal Service Level Agreement (SLA) was established in 2016 and has since been periodically updated and recently extended until June 2023. The SLA between EGI and WeNMR is a document summarizing ten Operational Level Agreements (OLAs) between EGI and ten resource centers committed to providing HTC, cloud compute and storage capacity with availability and reliability above well-defined targets constantly monitored by the EGI operation team. These ten resource centers are located in Czech Republic, Italy, Portugal, Spain, Taiwan, Netherlands, and Ukraine. They ensure the availability of 6050 HEPSPEC′06 normalized CPU-years per year of HTC computing time, 540 virtual CPU-cores of cloud compute capacity, and 59 TB of online storage capacity. Furthermore, a grid site hosted at the Consorzio Interuniversitario Risonanze Magnetiche di Metalloproteine Paramagnetiche (CIRMMP) provides several GPGPU cards that greatly enhanced the performance of the DISVIS, POWERFIT, and AMPS-NMR applications. Next to the SLA sites, additional resources are also available on an opportunistic use basis. During 2020 the HTC capacity has further increased by adding some Open Science Grid resources hosted in the US and a few WLCG sites hosted in France, Germany and Spain that decided to support WeNMR COVID-19 related jobs.

The WeNMR community has since the beginning recognized the importance of having a Single Sign-On (SSO) mechanism to access their web portals with unique login credentials. The first SSO was implemented in 2013 under the WeNMR European project, allowing users to subscribe to portals, validate personal grid certificates and manage job submissions all from one web page, collecting at the same time accounting records and user’s geographic information. A few years later, with the support of the West-Life H2020 European project (2015–2018), the WeNMR SSO was gradually replaced with the West-Life SSO. In recent years the SSO mechanisms typically have relied on an Authentication and Authorization Infrastructure (AAI) based on services like the IdP-SP-Proxy and standard protocols like SAML, OpeniD Connect and OAuth, and have been designed following the guidelines of the Authentication and Authorization for Research and Collaboration (AARC) initiative launched in May 2015. With the support of the EOSC-Hub project (2018–2021) the WeNMR services transitioned to the current AAI solution based on the EGI Check-in service. Developed from the same AARC Blueprint Architecture ¹ , EGI Check-in solution is part of EOSC service catalogue, and has also been adopted by the Instruct-ERIC infrastructure. Nowadays all the WeNMR portals integrate EGI Check-in as SSO mechanism. Users can register to the EGI Check-in at https://aai.egi.eu and use those credentials to register to WeNMR services. All portals are compliant with the European General Data Protection Regulation (GDPR) with clear condition of use (https://wenmr.science.uu.nl/conditions) and privacy measures (https://wenmr.science.uu.nl/privacy).

The EGI Workload Manager is a service enabling users to access distributed computing resources of various types through a single user-friendly interface. It is one of the services offered by the EGI e-infrastructure project and it is also available via the EOSC Compute Platform. The service is built upon the software from the DIRAC Interware project (http://diracgrid.org) (Tsaregorodtsev and Project, 2014). The project is providing a development framework and a set of ready-to-use components to build distributed computing systems of arbitrary complexity. It provides a complete solution for communities needing access to computing and storage resources distributed geographically, integrated in different grid and cloud infrastructures or standalone computing clusters and supercomputers. The DIRAC Workload Manager serves multiple scientific communities–partners of the EGI e-infrastructure and the WeNMR Collaboration as one of its most active users. User tasks prepared by the WeNMR application portals are submitted to the Workload Manager service. The service performs reservation of computing resources by means of so-called pilot jobs which are submitted to various computing centers with appropriate access protocols. Once deployed on worker nodes, pilot jobs verify the execution environment and then request user payloads from the central DIRAC Task Queue. Altogether, the pilot jobs and the central Task Queue form a dynamic virtual batch system that overcomes the heterogeneity of the underlying computing infrastructures. This workload scheduling architecture allows to easily add new resources transparently for the users. It also makes it easy to apply usage policies by defining fine grained priorities to certain activities. For example, during the COVID pandemic it allowed to quickly make available resources of the sites willing to contribute to the COVID-related studies and ensure high priority of these jobs compared to other regular WeNMR activities.