Hungary's genomic data, analysed across institutions - without a record moving

Biobanks at Hungary's leading research universities hold genomic sequences and clinical phenotype data on thousands of patients raw material of enormous value for personalised medicine, drug target identification, and public health research. The problem is structural. This data is held in siloed institutional databases that do not communicate with each other, and it is too sensitive to combine through any traditional route: centralising it in a shared database, uploading it to a cloud platform, or granting external researchers direct access are all legally blocked under data protection regulation. The result is a paradox that characterises biobank research across Europe: the scientific potential of the data grows with every new sample added, yet it remains almost entirely unusable for the multi-site, large-cohort studies that actually require it. Research programmes that need diverse genomic populations must work with single-institution datasets too small to draw reliable conclusions or abandon the question.

E-Group, Semmelweis University, and Budapest University of Technology and Economics are jointly developing FedXGen, a new module within the FedX platform, purpose-built for genomic and phenotypic data analysis. FedXGen extends federated learning into the specific demands of genomic research, incorporating two classes of AI that represent the current frontier of the field: polygenic risk modelling, which combines signals from thousands of genetic variants across the genome to predict individual disease susceptibility; and transformer-based foundation models, the architectural breakthrough behind modern large language AI, here applied to the structured language of genomics and clinical phenotypes. The principle is the same as across all FedX deployments: models travel to the data, not the other way around. FedXGen sends its analytical processes to each biobank, runs them entirely within that institution's environment, and returns only anonymised, aggregated statistical outputs. No patient record moves. No institution loses control of its data. Alongside FedXGen, the project builds NEFAB (the National Federated Biobank Data Network) connecting Semmelweis University and the University of Szeged as its initial two nodes, creating Hungary's first infrastructure for federated joint analysis of genomic and phenotypic data across institutional boundaries.

NEFAB will be the first infrastructure in Hungary enabling federated genomic research across multiple institutions without moving personal data. Through FedXGen, researchers at Semmelweis and Szeged and, as the network grows, institutions beyond, will be able to run joint analyses on combined genomic cohorts: drug development studies, polygenic disease risk assessments, and precision treatment research that would have been legally and technically impossible before. The project targets a commercially mature FedXGen platform by May 2027, capable of forming self-sustaining federated biobank networks at the international level — in both research and commercial contexts — reducing the cost and time of genomic research programmes while increasing their statistical power and reliability. The longer-term ambition is explicit: connect NEFAB to Hungary's national healthcare system and position Hungary as a leading country in precision medicine internationally. FedXGen is also the genomics-specific extension of E-Group's broader federated AI strategy; the same platform logic that powers FedX deployments in clinical AI (HeliX), automotive (Bosch), and cloud-edge infrastructure (TIM, Virt8ra), now extended into the molecular layer of biomedical research.