How can AI help detect Parkinson’s disease? CU Anschutz shares HEREDITARY research

The University of Colorado Anschutz Medical Campus (CU Anschutz), HEREDITARY’s only U.S. partner, has published a new article showcasing its contribution to the project and the potential of AI to support the early detection of Parkinson’s disease through retinal imaging.

Within HEREDITARY, researchers at the Sue Anschutz-Rodgers Eye Center are investigating whether AI can identify retinal biomarkers associated with neurodegenerative diseases such as Parkinson’s disease and multiple sclerosis. By combining retinal images with other clinical and biomedical data, the team aims to develop multimodal AI models that could contribute to earlier, less invasive diagnosis.

The article also highlights the role of federated learning, a key approach within HEREDITARY that enables institutions to collaboratively train AI models while keeping sensitive patient data securely at their original locations. This privacy-preserving methodology allows researchers across the consortium to build more robust and reliable models without sharing raw clinical data.

As HEREDITARY’s only U.S.-based partner, CU Anschutz brings internationally recognised expertise in ophthalmic AI, oculomics, and federated learning, strengthening the consortium’s efforts to better understand the gut-brain axis and advance research into neurodegenerative diseases.

Want to learn more about CU Anschutz’s contribution to HEREDITARY? Read the full article on their website. READ NOW!

#DeCoding HEREDITARY: Building trustworthy AI for Health Data in Europe’s evolving regulatory landscape

Artificial Intelligence is transforming biomedical research, but innovation alone is not enough. To unlock the full potential of health data, researchers must also ensure that new technologies are developed responsibly, transparensively, and in full alignment with Europe’s evolving legal and ethical framework.

This is the focus of HEREDITARY Deliverable 7.2, an in-depth legal and ethical study that moves beyond theoretical compliance to analyse how European legislation applies to the project’s real-world research activities and federated architecture. Building on the foundations established in Deliverable 7.1, the study assesses how HEREDITARY’s technologies align with the latest European regulations while identifying the challenges that must still be addressed as the project progresses.

GDPR remains at the heart of HEREDITARY

One of the main conclusions of the study is that the General Data Protection Regulation (GDPR) continues to be the cornerstone governing the project’s processing of health and genetic data.

HEREDITARY’s federated approach supports several key GDPR principles, including data minimisation, privacy by design, and reducing the need to transfer sensitive information between institutions. By keeping patient data within each participating organisation, the project significantly reduces many traditional data-sharing risks.

However, federated learning is not automatically GDPR compliant. Even when raw data never leaves local institutions, model parameters, analytical outputs or derived inferences may still reveal personal information under certain circumstances. For this reason, GDPR compliance must be ensured throughout the entire data lifecycle, including data collection, local processing, federated model training, analytics, output generation, storage, and cross border collaboration.

Preparing for the European Health Data Space

Another major focus of Deliverable 7.2 is the European Health Data Space (EHDS), one of the EU’s flagship initiatives for enabling secure secondary use of health data across Europe.

The analysis concludes that HEREDITARY is already strongly aligned with many of the EHDS objectives. Its emphasis on interoperability, privacy-preserving data analysis and federated infrastructures positions the project as a promising contributor to future European health data ecosystems, including potential integration with HealthData@EU infrastructures.

AI Act: responsible AI goes beyond today’s requirements

Because HEREDITARY currently operates as a research infrastructure, many obligations introduced by the AI Act are likely to remain limited during the project’s research phase. Nevertheless, the Deliverable stresses that this situation is dynamic. If AI components developed within HEREDITARY eventually evolve into clinical decision-support tools or become deployed in operational healthcare settings, additional regulatory requirements related to risk management, transparency, human oversight and conformity assessment could become applicable.

This forward-looking assessment allows the consortium to anticipate future obligations long before technologies reach clinical practice.

DGA & NIS2: Governance and cybersecurity as key enablers

The Data Governance Act (DGA) reinforces principles that closely match HEREDITARY’s federated design, promoting secure processing environments, interoperability, responsible reuse of protected datasets and transparent governance mechanisms. Meanwhile, the NIS2 Directive highlights the growing importance of cybersecurity in distributed health data infrastructures.

Federated learning offers important privacy advantages, but it also introduces new types of cybersecurity risks, including model poisoning, inference attacks and parameter leakage. The study concludes that cybersecurity, data protection and governance must be addressed as an integrated framework, recognising that protecting health data is both a legal obligation and a prerequisite for trustworthy biomedical research.

Looking beyond compliance: ethics in federated AI

HEREDITARY combines genomics, artificial intelligence and cross-border health research. This makes ethical governance essential throughout the project. The study identifies ongoing challenges related to fairness, transparency, explainability, accountability, informed participation and responsible stewardship of research data.

Federated architectures substantially reduce the need for centralised data sharing, but they do not eliminate ethical risks associated with bias, re-identification, opaque AI models or the downstream use of research outputs. These issues require continuous assessment as both technologies and regulations evolve.

Bringing regulation closer to real-world research

One of the most valuable contributions of Deliverable 7.2 is its Use Case-Specific Analysis, presented in Annex 3.

Instead of relying on abstract legal interpretation, WP7 evaluates how European regulations apply to the concrete research scenarios developed across HEREDITARY’s five clinical use cases. It connects legal analysis directly with real technical workflows. This practical assessment also addresses one of the consortium’s previously identified governance gaps: situations where one partner’s analytical tools are applied to another partner’s datasets within the federated infrastructure.

These findings will now guide the final phase of Work Package 7, where the consortium will identify the remaining regulatory and ethical gaps and develop concrete recommendations to ensure that the HEREDITARY framework remains not only innovative, but also compliant, trustworthy and sustainable for future healthcare research across Europe.

#DeCoding FAIRification: making health data discoverable without compromising privacy

Imagine two hospitals collecting information about patients with the same neurological disease. Both have valuable data for researchers, perhaps to find a better treatment. Both are willing to collaborate with researchers. And yet, nobody outside these hospitals know exactly what data exist or whether they could help them in their research of the disease. As a result, these two potentially valuable datasets remain invisible, even though they could contribute to improving the life of the patients. 

This is one of the less visible, but most important, challenges in modern biomedical research.Before researchers can analyse data, they first need to know that the data exist. That is where FAIRification comes in. Check out the following video, in which Marcos Casado, Senior Metadata Bioinformatician at EMBL-EBI, explains it:

Within HEREDITARY, Work Package 3 is not only building the technological infrastructure for federated analytics and semantic interoperability. It is also ensuring that the project’s data resources become easier to find, understand and reuse, while always respecting privacy, institutional ownership and legal requirements. This work was included in Deliverable D3.6, FAIRification of Participating Data Resources, and the successful achievement of Milestone 7 at the end of 2025. 

FAIR does not mean open. Sensitive clinical and genomic data often cannot leave the institutions that generated them, due to legal and ethical constraints. HEREDITARY’s proposed FAIRification focuses on making datasets discoverable, before access to sensitive data is even considered. This way, researchers can understand what information exists, under which conditions it may be accessed and how it relates to other resources, without exposing any sensitive information. In essence, it promotes collaboration across institutions while respecting the aforementioned constraints. 

Making this possible begins with something that may seem surprisingly simple: describing data better. Every dataset carries information beyond the measurements themselves, such as how it was generated, what variables it contains, which standards were used or under which conditions it can be shared. This information, known as metadata, acts as a guide for both researchers and computers. HEREDITARY works to harmonise these descriptions using internationally recognised standards and controlled vocabularies, creating a common language that allows clinical, genomic and imaging datasets to be understood consistently across institutions and countries. 

Beyond the project itself, FAIRification also prepares HEREDITARY for future European collaboration. To achieve this, the project aligns its metadata with internationally recognised frameworks and recommendations, including those promoted by the 1+ Million Genomes initiative (1+MG), the Global Alliance for Genomics and Health (GA4GH) and the European Health Data Space (EHDS). These shared standards ensure that today’s datasets can continue supporting tomorrow’s research. In this way, HEREDITARY is not only developing new technologies, but also contributing to a more connected, interoperable and trustworthy European research ecosystem. 

Over the past two years, the project has assessed participating datasets and implemented practical FAIRification workflows adapted to different situations. Some of the results presented in Deliverable 3.6 are: 

  • Richer metadata descriptions for consortium datasets.
  • Harmonisation of metadata across participating institutions.
  • Preparation of selected datasets for deposition in the European Genome-phenome Archive (EGA).
  • Publication of metadata records following European HealthDCAT-AP recommendations.
  • Reusable FAIRification workflows for future datasets.
  • Quality assessment procedures to monitor improvements over time.  

If you would like to explore FAIRification in more detail, check out how FAIRification moves from theory to real-world implementation in biomedical research: 

FAIRification is about ensuring that valuable health data remain discoverable, understandable and reusable while fully complying with the General Data Protection Regulation (GDPR) and respecting the autonomy of the institutions that generate them. By building on internationally recognised standards, trusted repositories and established European infrastructures, HEREDITARY is creating the conditions for future scientific collaboration that extends well beyond the lifetime of the project. Combined with semantic interoperability, federated analytics and privacy-preserving infrastructures, FAIRification forms another essential pillar of HEREDITARY’s vision: enabling a more connected, trustworthy and collaborative ecosystem for biomedical research across Europe. 

New HEREDITARY results bring Europe closer to secure, federated AI for health research

The HEREDITARY project has reached another important milestone in its third year by delivering a new set of scientific and technical results to the European Commission. At Month 30, the consortium has completed six key deliverables that reinforce the project’s vision of enabling secure, privacy-preserving and multimodal research on neurodegenerative diseases across Europe.

These results represent significant progress across several work packages, ranging from clinical use cases and federated infrastructures to artificial intelligence, legal and ethical frameworks, and privacy-preserving analytics.

From data preparation to federated multimodal ALS research

One of the most relevant achievements is the publication of D2.17: Neurodegenerative Use Cases: Intermediate Results, led by the University of Torino (UNITO). This deliverable marks an important step towards federated multimodal research on neurodegenerative diseases. Using Amyotrophic Lateral Sclerosis (ALS) as its main demonstrator, the deliverable shows how HEREDITARY is moving from use-case design to operational data readiness, integrating FAIRified datasets, semantic interoperability through the HERO ontology, privacy-preserving genomic discovery, and biological insights into a common framework for federated research. Another important aspect of this work is the systematic inclusion of sex-disaggregated analyses, helping identify biological differences that may influence disease progression, diagnosis or treatment response.

The deliverable also outlines the project’s roadmap for ontology-enabled multimodal patient stratification, combining advanced machine learning and semantic technologies to support future federated analyses across institutions. In doing so, it validates Milestone 9 and establishes the scientific and technical foundations for the next phase of HEREDITARY’s neurodegenerative use cases.

Building trustworthy federated AI through privacy-preserving analytics

The newly released D3.8: Privacy-preserving Analytics: First Release, coordinated by Aalborg University (AAU), presents the first implementation of the privacy-preserving technologies that will be integrated into HEREDITARY’s federated analytics and machine learning platform.

The report describes how multiple complementary privacy technologies, including differential privacy, secure aggregation and homomorphic encryption, can be combined to protect sensitive information while still enabling collaborative data analysis across institutions. Beyond describing the underlying algorithms, the deliverable introduces mechanisms to continuously monitor privacy risks during federated computations and evaluates how different protection levels can be adapted depending on the analytical workflow.

These developments establish the privacy layer that will support future federated AI models and formally achieve Milestone MS10, dedicated to privacy-preserving methods within HEREDITARY.

Combining biomedical data for more accurate AI models

The publication of D4.5: Multimodal Learning Methods, led by Radboud University Medical Center (RUMC), presents a new set of computational tools and machine learning methods for multimodal biomedical analysis focused on gut health.

The deliverable also evaluates different strategies for combining histopathology, microbiome and clinical data. The results show that integrating histopathology with fecal microbiome data provides the strongest predictive performance, establishing an important methodological baseline for future multimodal analyses within HEREDITARY. The deliverable also foresees the incremental publication of trained models through Grand Challenge platforms and as open-source software, supporting transparency, reproducibility and collaboration within the scientific community.

Strengthening ethical and legal foundations for European health data sharing

The consortium has released D7.2: In-depth Legal and Ethical Study, coordinated by KU Leuven, which expands the initial legal and ethical inventory developed earlier in the project.

The report analyses the regulatory challenges associated with federated research infrastructures, including GDPR compliance, secondary use of health data, governance responsibilities and cross-organisational collaboration. Rather than providing only theoretical guidance, the study evaluates these issues across HEREDITARY’s clinical use cases, offering practical recommendations for implementing secure “tool-over-data” approaches where algorithms travel to the data instead of transferring sensitive datasets.

Complementing the legal study, the consortium has also published the updated D2.2: Ethical Guidelines, Data Collection and Sharing. The document revises the project’s ethical guidance based on the experience gained during the first half of the project and reflects the evolving requirements for federated model training and multicentre collaboration. It updates recommendations for data sharing, incorporates revised procedures for prospective data collection where necessary and strengthens the links between ethical governance activities and the project’s wider regulatory framework.

Federated infrastructure Implementation

In parallel, the consortium has successfully submitted D2.12: Federated Infrastructure Implementation, marking another important technical achievement for the project. This deliverable advances the implementation of HEREDITARY’s federated infrastructure, which enables secure distributed analysis across participating institutions without centralising sensitive data. Although the deliverable has been officially delivered to the European Commission, it is not yet publicly available because parts of its content are currently under scientific review and publication.

Looking ahead

These new deliverables illustrate how HEREDITARY continues to transform its scientific vision into practical technologies and validated methodologies.

By simultaneously advancing federated infrastructures, privacy-preserving AI, multimodal machine learning, integrated clinical research and trustworthy governance frameworks, the project is steadily building an ecosystem where researchers can collaborate across Europe while keeping sensitive health data protected. As the project moves towards its next phase, these achievements provide a solid foundation for the forthcoming demonstrations, validation activities and clinical applications that will further accelerate innovation in neurodegenerative disease research.

#DeCoding HEREDITARY: making health data understandable through visual analytics

One of the biggest challenges in HEREDITARY is not only to collect and secure integrate data, but also to make sense of it. 

Researchers, clinicians, policymakers and citizens are increasingly confronted with vast amounts of information coming from medical images, genetic data, microbiome profiles, electronic health records, simulations and many other sources. While these datasets hold enormous potential to advance our understanding of health and disease, their complexity can make them difficult to interpret and use effectively. 

This is where HEREDITARY’s Work Package 5 (WP5), coordinated by TU Graz, comes in. Through the development of innovative visual analytics methods and interactive exploration tools, WP5 helps transform complex multimodal data into understandable insights that can support research, prevention and decision-making across the healthcare ecosystem. 

Today, we are excited to showcase a big result coming from this work: the launch of the HEREDITARY Demos & Visualisation Components Portal, publicly available at: https://demos.hereditary-project.eu/.

 

From research prototypes to publicly accessible demonstrators 

Over the last two years, WP5 has progressively transformed visualisation concepts into operational demonstrators and interactive applications. 

The developments reported in a series of deliverables (D5.1D5.2D5.3 & D5.4) include visualisation components for: 

  • High-dimensional biomedical data.
  • Knowledge graphs and semantic resources.
  • Brain imaging and spatial data.
  • Time-series and biosignal analysis.
  • Simulation and modelling outputs.
  • Natural language-assisted visual analytics. 

A key principle throughout this work has been openness and reusability. To make these developments accessible to a broader audience, TU Graz has established a dedicated demonstrator infrastructure that hosts and deploys visual analytics applications developed within HEREDITARY. The new Demos & Visualisation Components Portal now brings many of these innovations together in a single public entry point. 

Created through close collaboration between TU Graz and partners across the consortium, including experts in medical research, federated infrastructures, machine learning, data management and semantic technologies, the portal demonstrates how advanced visual analytics can support the exploration of multimodal health data. The portal currently includes 15 demonstrators, videos (in some cases) and code (available in most of them), from semantic exploration and cohort analysis to brain imaging, machine learning interpretation and simulation-based research.

 

Exploring the gut-brain connection through visual analytics 

Among the flagship developments showcased in the portal is the Gut Brain Explorer, an advanced visual analytics application designed to explore relationships between gut microbiota and brain activity. 

The tool combines multiple linked visualisations to allow researchers to investigate outputs generated through Linked Independent Component Analysis (LICA), integrating microbiome information with functional brain imaging data. Users can interactively explore microbiota distributions, modality contributions and brain activity patterns through coordinated views. 

The component has already demonstrated its scientific value during project evaluations, supporting researchers in identifying biologically relevant gut-brain associations.

 

Making complex biomedical data easier to explore 

Several other demonstrators address complementary challenges in data exploration and interpretation. 

Clusters in Focus helps researchers identify and compare meaningful patient subgroups within high-dimensional biomedical datasets, supporting tasks such as biomarker discovery and phenotyping. 

Neurodegen-Vis combines interactive visual analytics with LLM-powered assistance to support the exploration of healthcare datasets. The tool enables users to investigate correlations and dependencies in medical data while receiving guidance through natural language interaction. Privacy-preserving mechanisms are integrated to protect sensitive information. 

OnSET (Ontology and Semantic Exploration Toolkit) helps users navigate complex knowledge graphs and ontologies through natural language querying and visual graph exploration, making semantic resources more accessible to non-experts.

 

Building trust through transparency and interaction 

One of HEREDITARY’s core ambitions is to ensure that advanced AI and data-driven methods remain understandable and trustworthy for the people who use them. Visualisation plays a crucial role in achieving this goal. 

By allowing users to interact directly with data, inspect results, understand relationships and explore evidence behind conclusions, visual analytics can help make complex technologies more transparent and interpretable. This is particularly important in healthcare, where trust, explainability and human oversight remain essential. 

As HEREDITARY progresses, new demonstrators and functionalities will continue to be added, further expanding the ecosystem of tools available for exploring multimodal biomedical data, semantic resources and AI-driven analyses. 

🔗 Explore all the demonstrators and get in touch with the team responsible for each one: https://demos.hereditary-project.eu/ 

#DeCoding Federated Analytics: unlocking knowledge across borders and datasets

How can researchers study complex diseases across Europe when sensitive and health data cannot leave hospitals or research centres and speaks several languages?

This is one of the key challenges HEREDITARY is addressing, and part of the answer lies in a powerful combination of a federated learning infrastructure, semantic integration and federated analytics.

In this new #DeCoding article, we take a closer look at how Work Package 3 (WP3) is building the foundations that make this possible: the Hereditary Ontology (HERO) and the Hereditary Data Network (HDN).

 

From federated learning to federated analytics

In a previous #DeCoding article, we explored federated learning, a method that allows AI models to be trained across multiple institutions without centralising raw data. But before researchers can analyse data across institutions, they need to ensure they are actually talking about the same things. In healthcare, the same clinical concept can be recorded differently depending on the hospital, the specialty, or even the country. This makes it difficult to combine or compare data.

To solve this, HEREDITARY has developed the Hereditary Ontology (HERO): a shared semantic layer that provides a common language for all partners. This allows different datasets to be understood in a consistent way. It enables researchers to formulate questions without needing to know local database structures, by integrating clinical, genomic and imaging data into a unified conceptual model, covering key neurological diseases domains, such as Amyotrophic lateral sclerosis (ALS) and Multiple sclerosis (MS), and designed to expand to others like Parkinson’s and Alzheimer’s.

This semantic integration is essential: without it, federated analytics wouldn’t be possible.

 

From data silos to a connected network

Building on this ontology, HEREDITARY has developed the Hereditary Data Network (HDN), a federated infrastructure that allows data to be analysed across institutions while remaining locally stored. Data stays where it is, but knowledge can travel. Instead of moving individual patient data to a central repository, HDN enables researchers to send queries to different institutions and receive aggregated results. It is based on a central component that coordinates queries, local endpoints at each institution that execute them on their own data and results that are returned and combined, without exposing sensitive information.

This approach represents a fully federated and privacy-by-design architecture. Privacy controls are integrated in the query processing layer of HDN:

  • Each query is assessed before running and it is automatically assigned with a privacy risk score.
  • Each institution decides what are the risk thresholds they can safely handle.
  • If a query exceeds that thresholds, no data is returned or privacy mitigation measures are applied.

This ensures that data owners remain in full control, while still enabling meaningful research across institutions.

 

How federated analytics works in practice? 

A researcher might ask a question like: “What is the average age at onset of ALS patients?”. 

Instead of accessing a central database, the system: 

  1. Translates the question into a standardised query using HERO.
  2. Sends it to multiple institutions.
  3. Executes it locally at each site.
  4. Returns aggregated results.
  5. Combines them into a single answer, obtaining a response that incorporate insights across different datasets while respecting privacy and institutional autonomy. 

 

Progress so far and what comes next 

By now, HEREDITARY has already made significant progress. The project has delivered the first version of its federated workflow execution methods (D3.2) and demonstrated how semantic integration and federated querying can work together. Also, the HDN prototype has shown that distributed queries can be executed across heterogeneous datasets, integrating privacy-aware query mechanisms. For those with a technical interest, various resources relating to these developments can be found on the project’s Open Hub. 

Looking ahead, the project is focusing on scaling and real-world deployment. Over the first half of 2026, HDN endpoints are being installed across several partners (University of TurinRadboud University Medical Centre and University of Colorado), enabling future live queries on real datasets. The goal is to have a fully operational federated query system running at consortium level by the end of 2026, along with a shared catalogue of queries and a clear maintenance plan.  

Ultimately, what HEREDITARY is building goes beyond technology. It is a new way of doing research in several fields: one where data does not need to move to generate knowledge, where institutions can collaborate without losing control and privacy, and where complexity is managed through shared understanding. The Federated analytics layer, powered by HERO and the HDN, is a key step in that direction.

 

Learn more about Federated Analytics in the following videos, where our coordinator, Gianmaria Silvello (University of Padova) and Daniele Dell’Aglio (Aalborg University) share their insights and perspectives on the topic: