Loading...
Date
2025
Abstract
The exponential growth in health data, driven by innovations in experimental methods and new data sources, has rendered manual analysis intractable in many cases. This intractability necessitates the use of AI tools and methods to automate previously unfeasible analysis tasks. This work addresses the challenge of enabling health domain researchers to effectively utilize AI methods for extracting actionable insights from raw data, while overcoming the barriers posed by the technical complexity of these tools.
The interdisciplinary nature of modern health research, coupled with the rapid innovation in ML-based analysis algorithms, has created a significant challenge: the technical overhead required to implement these computational tools often excludes a large body of researchers from participation. This exclusion potentially slows the rate of advancement in health research by limiting the pool of researchers who can actively participate. Moreover, the integration of independently developed AI tools, which often lack interoperability and have complex dependencies, presents significant challenges in creating cohesive analysis workflows.
Current approaches require researchers to possess programming expertise, diverting their focus from primary research objectives to acquiring computational skills. The use of general-purpose programming languages not only creates a steep learning curve but also impacts researchers’ productivity and experimental output. Furthermore, existing LCNC methods lack semantic type-based validation of data flow, potentially leading to the misapplication of ML models. Also as ML models do not have a guarantee of correctness, the absence of comprehensive runtime intervention capabilities, to inspect intermediate results and if necessary provide additional input raise concerns about result interpretation and validity. Similarly in the case of ad-hoc data processing workflows developed by domain researchers there is limited data provenance tracking, raising further concerns about reproducibility, the cornerstone of modern science.
To address these challenges, this work proposes a LCNC platform for AI-driven health analysis, introducing a novel architecture and ontology-driven DSL-based approach for tailoring the general purpose workflow architecture to specific target domains. The platform incorporates semantic type systems which enables semantic type-based validation of data flow, supports runtime feedback, and can enhance the reliability and interpretability of AI workflows. By leveraging MDD principles and semantic web techniques, the proposed solution aims to bridge the “semantic gap” between health domain researchers and AI technologies.
The methodology was applied in two distinct health domains: public health (colon cancer information veracity analysis) and biomedical research (highly-plexed immunofluorescence image analysis). These implementations demonstrate the platform’s versatility and effectiveness across different health research contexts. The platform offers a cohesive interface for creating AI analysis workflows with heterogeneous components, providing an intuitive means for health domain researchers with limited programming expertise to leverage state-of-the-art AI methods.
This research contributes to addressing key questions in the field, including how ontology-based semantic associations can improve the
adoption of AI methods in health research, how static compatibility checks can prevent the misapplication of domain-specific AI technologies, and how MDD can enhance interdisciplinary collaboration in both health AI workflow design and tool development.
By facilitating the integration of domain-specific knowledge with AI technologies, this approach can not only bridge the gap between health research and AI but also promote reproducibility, enhance interdisciplinary collaboration, and accelerates the adoption of AI in advancing health research.
Supervisor
Margaria, Tiziana
Description
Publisher
University of Limerick
Citation
Files
Loading...
Brandon_2025_No-Code.pdf
Adobe PDF, 46.98 MB
ULRR Identifiers
Funding code
Funding Information
Sustainable Development Goals
External Link
Type
Thesis
Rights
http://creativecommons.org/licenses/by-nc-sa/4.0/