The project “Extension of the Risk Assessment Knowledge Integration Platform RAKIP-Web” was carried out as an
EFSAshort forEuropean Food Safety Authority tailor-made activity in the area “Data”. The starting point was the following practical problem in the field of risk assessment: data, models and model code are often created in individual projects, but are not documented and made available in such a way that they can later be understood by other experts, re-executed or used in other contexts. This particularly affects quantitative assessment models from the field of food safety and in the One Health context, such as microbiological risk assessments, dose-response relationships, risk-benefit assessments, predictive microbiology or epidemiological models. The FSKX format developed by RAKIP-Initiative addresses this problem by bringing together model code, input data, execution scenarios and metadata in a standardized format. With the RAKIP-Web web portal,
BfRshort forGerman Federal Institute for Risk Assessment, in cooperation with the members of RAKIP-Initiative, provides a web-based environment in which such FSKX models can be stored, curated, found and, in some cases, executed directly online. The aim of the RAKIP-Web project was therefore to facilitate the use of FSKX and RAKIP-Web, to modernize the technical infrastructure, to add new models to the RAKIP-Web model repository and to create a robust foundation for future development. The four project deliverables corresponded to these target areas: development of an online course with video tutorials and webinars, extension of relevant software tools, inclusion of additional risk assessment models and preparation of a roadmap for future maintenance and further development. In addition, a network of FSKX stewards was to be established or strengthened, i.e. persons who support model authors in the creation, documentation and curation of FSKX files. A central component of the project was capacity building. The online course developed is aimed at risk assessors, modellers and experts from authorities, research institutions and
EFSAshort forEuropean Food Safety Authority bodies. The training materials first explain the basic idea of RAKIP-Web and demonstrate how existing models in the repository can be found, inspected and executed. Then, using a simple example, it is shown how a Python model is converted into the FSKX format: model parameters are identified, metadata are derived from a publication and the model description is supplemented in such a way that later reuse becomes possible. Further modules deal with the automatic validation of an FSKX file, including a standard simulation, as well as with the curation process and the quality aspects examined in this context. This conveys both the technical use of the platform and the content-related requirements for models that can be understood and reviewed. In addition, three webinars on RAKIP-Web, FSKX Cloud Platform and model curation were held. The feedback from these was incorporated into the course structure. The course has been prepared for provision via EU Academy and can in future be used systematically for the onboarding of new FSKX stewards. The technical developments focused on three areas: support for FSKX in the programming language R, the semantic description of models and the web-based execution of more complex model types. In the course of the project, the R library “FSK2R” was updated to version 0.2 and adapted to the current FSKX metadata schema. As a result, newer FSKX models, for example from RAKIP-Web or EFSA’s Knowledge Junction, can be read, edited, executed and re-exported in R. Improvements include the recognition of script names from the metadata, more robust handling of working directories and referenced files, and a consistent cross-platform treatment of text and script files. This lowers entry barriers for modellers and supports reproducible analyses. In the second work area of this work package, an FSKX ontology (FSKXO) was developed to improve the interoperability of models. Ontologies describe technical terms and their relationships to one another in a machine-readable form and thus make it possible to interpret metadata in a uniform way. In the project, all 174 fields of the FSKX metadata schema were mapped to existing public ontologies or, if no suitable equivalent was available, included as FSKX-specific properties. In addition, 49 controlled vocabularies were integrated and documented with standardized mapping information. Building on this, an ETL pipeline was set up that converts existing metadata.json files into JSON-LD. JSON-LD is a linked data format with which metadata can be better interpreted by both humans and software and later used in knowledge graphs. This laid the foundation for semantic search and for a future RAKIP knowledge graph in which model parameters and relationships can be queried on a graph basis. The third technical focus was the execution of models in the cloud. In the FSKX-Web platform developed as a prototype in the KIDA project, specific containers can be provided for individual models that contain the software required in each case. These containers are operated in a Kubernetes environment as needed, which enables scalable and reproducible execution. This makes it possible to run models that require additional software such as JAGS, Stan or BUGS, as is often used in Bayesian Monte Carlo models. These functionalities were tried out and tested by the project partners using several example models, including models for the modelling of mixture distributions, population sizes, zero-inflated models, time series models and dose-response models. The FSKX-Web approach reduces dependencies on local installations and supports traceable execution scenarios across different institutions. The RAKIP-Web model repository established on the FSKX-Webplatform was also expanded in the project to include a number of new model types. The original target number for the project was to include at least three new risk assessment models; in fact, significantly more models were integrated into the new RAKIP-Web model repository by the project partners and by
BfRshort forGerman Federal Institute for Risk Assessment. These include the probabilistic sQMRA model for rapid microbiological risk assessments, components from the TOXOSOURCES model on
Toxoplasma gondii, a Danish Toxoplasma QMRA, a model derived from AirCoV on airborne virus transmission indoors, a risk-benefit model for the substitution of meat with pulses, models on TBEV infections through tick bites, a beta-Poisson dose-response model for
Mycobacterium bovis, a time series model for histamine surveillance and an FSKX module for the
EFSAshort forEuropean Food Safety Authority model on the persistence of
Listeria monocytogenes in processing environments. Twelve models of different types were converted from MicroHibro into FSKX, including growth, inactivation, growth boundary and dose-response models, which can now be visualized and executed in standardized form. A special effort was undertaken on the integration of model metadata within the existing openFSMR model inventory, which is extensive directory of predictive microbiological models. The openFSMR directory was first curated, updated and cleared of obsolete entries. The columns of the openFSMR data set were then mapped to FSKX metadata concepts. An automatic pipeline generated model-specific FSKX metadata in JSON-LD format from this information, supplemented missing mandatory information, tested MIRARAM conformity and bundled the results into model-specific FSKX files. These were made available in a dedicated openFSMR model repository on the FSKX-Web platform. In addition, an AI-supported mapping approach was tested to link terms from openFSMR with public ontologies and thesauri, such as taxonomies for microorganisms. Expert-based performance evaliuation showed an overall performance of 88%. The results show that large model inventories can technically be transferred into the FSKX ecosystem, even if quality assurance and expert review remain necessary for this. The work from the deliverables jointly contributes to the FAIRness of the model resources. Models become more findable because structured metadata, controlled vocabularies and, in perspective, semantic search functions can be used. They become more accessible because RAKIP-Web,
EFSAshort forEuropean Food Safety Authority Knowledge Junction, Zenodo and web-based services serve as distribution and access points. Interoperability is improved through FSKX, FSKXO, JSON-LD, FoodEx2 references and ontology mappings. For the One Health area, it is particularly relevant that the repository not only includes classical food-related models, but also models on vector transmission, indoor , epidemiological time series, environmental persistence and nutrition-related risk-benefit questions. At the same time, the project revealed several limitations. The FSKX metadata schema is extensive and can cause a high documentation effort for model authors. For complex models, complete transfer into a single executable FSKX package is not always immediately possible; for example, the complete TOXOSOURCES main model could not be integrated due to its dependence on several submodels, so that individual components and an alternative Toxoplasma model were provided instead. In the AirCoV-based model, the basic functionality was implemented in R, while certain probabilistic scenarios of the original Mathematica version were initially not included. In the fourth work package, analyses and concepts for the long-term continuation of the RAKIP-Web resources were carried out and developed. The roadmap prepared in this work package takes up the aspects relevant for sustainability and makes concrete proposals for implementation in the period from 2025 to 2027. For example, the FSKX format is to be simplified and supplemented by a clearly separated execution profile in order to record runtime information (including containers) in a more structured way. Linked data, JSON-LD and ontology integration are to be further anchored in the software landscape so that a RAKIP knowledge graph can be built and used for extended search and comparison functions. AI-supported functions are to be tested and validated, for example for the creation of FSKX files, the extraction of information from literature and opinions, support for curation, checking MIRARAM conformity, searching for models and interpreting simulation results; IT security and the requirements of the EU AI Act are to be taken into account from the outset. In parallel, cooperation with standardization initiatives and projects such as FoodGuard, Ambrosia and MicRISK2030 is to be expanded and the steward community strengthened through training materials, events and a simplified legal framework. Overall, all deliverables defined in the project proposal were fulfilled. The online course and the webinars support the use of RAKIP-Web and FSKX and contribute to the development of a steward community. The software work improved the practical creation, editing, semantic description and cloud execution of FSKX models. The model inventory was expanded far beyond the minimum requirement and covers several risk assessment domains along the One Health chain. The roadmap describes concrete next steps for maintenance, standardization, AI support, community building and governance. The long-term benefit of the project lies in the combination of standard format, repository, execution environment, curation and training. This combination can help to ensure that risk assessment models are documented more transparently, can be reviewed more easily, are networked in a more interoperable way and can be better reused in the future.