Introduction
In medical technology, the design of electromagnetic (EM) diagnostic and therapeutic tools—such as microwave ablation for tumor treatment or advanced cancer imaging—relies heavily on precise data regarding how biological tissues interact with EM fields. These interactions are characterized by dielectric properties, specifically relative permittivity and conductivity. Despite their importance, a significant "data gap" exists in this field. Experimental results are frequently confined to physical lab books, non-standardized spreadsheets, or isolated storage, often lacking the critical context (confounders) necessary for accurate comparison or reuse. Without this metadata, researchers are often forced to "reinvent the wheel," significantly slowing medical innovation.
To address this, researchers have developed a semantic framework designed to make dielectric measurements of biological tissues Findable, Accessible, Interoperable, and Reusable (FAIR). By implementing a custom metadata model and an automated "RDFization" pipeline, siloed laboratory data is transformed into a searchable, machine-readable resource.
Methods
The developed pipeline transitions data from semi-structured formats to a FAIR-compliant architecture. This process integrates several key semantic web technologies:
- MINDER Standard: The Minimum Information Model for Dielectric Measurements of Biological Tissues (MINDER) provides a reporting framework that ensures essential experimental confounders—such as tissue temperature, excision time, and hydration levels—are recorded.
- RDFization Pipeline: A Java application utilizing the Apache Jena framework parses semi-structured Excel data and transforms it into Resource Description Framework (RDF) triples, establishing meaningful relationships between data points.
- Ontology Integration: To ensure interoperability, the system reuses terms from established life science ontologies (found via tools like Ontobee and BioPortal), allowing the data to speak the same "language" as other global life science databases.
- Persistent Identifiers (PIDs): Unique, permanent URLs are assigned to every data object using identifiers.org and the MIRIAM registry to prevent link rot and ensure long-term accessibility.
The infrastructure is hosted on AWS, utilizing a Virtuoso triple store for data management, an Nginx web server, and a PHP-based frontend for human interaction.
Results and Discussion
The primary output of this framework is the Bio-MINDER database, a repository containing measurements for various tissues, including porcine kidney, bladder, and liver. The system's efficacy was validated through "competency questions"—complex, multi-variable queries that actual data consumers might pose, such as searching for liver data within specific temperature and frequency ranges. The system successfully answered these queries via its SPARQL endpoint.
A core insight of this work is that raw dielectric numbers (permittivity and conductivity) are insufficient without comprehensive metadata. By capturing clinical confounders like tissue state (in-vivo vs. ex-vivo), the data becomes scientifically robust for secondary analysis. Furthermore, the dual-access approach—providing a web portal for researchers and a SPARQL endpoint for machines—lowers the barrier to entry for both bench biologists and data scientists, moving data from an isolated "silo" to a "node" in the global data web.
Limitations and Future Work
While this first iteration successfully demonstrates the "FAIR-ification" of dielectric data, several areas for improvement remain. Currently, data licensing is not yet in a machine-interpretable format (e.g., Creative Commons RDF tags), which limits fully automated AI reuse. Additionally, the vocabulary matching system is currently restricted to exact matches; future updates will include broader term matching to enhance discoverability.
The current scope is focused on dielectric measurements, but the underlying pipeline is adaptable to other data-intensive fields. Future versions may also incorporate more sophisticated access controls for commercially sensitive or restricted datasets.
Summary updated