MSc Research Project · City St George's, University of London · 2024–2025 Problem Large-scale biomedical ontologies (DOID, NCIT, FMA) don't speak the same language. Traditional lexical matching methods miss thousands of valid concept mappings, creating interoperability gaps in medical data systems. Solution A two-phase alignment pipeline combining LogMapBio for initial mapping with SBERT transformer models for semantic validation — catching mappings that pure string-matching misses. Results
92.48% maximum recall achieved 826+ correct mappings recovered that lexical methods missed 70% reduction in processing time via parallel execution 38 minutes total runtime across 24,000+ entities
Tech Stack Java · LogMapBio · SBERT · Python · SPARQL · BioPortal API Key Innovation A "Merged Strategy" combining lexical and semantic approaches, outperforming either method used alone.
- LogMap - Dependencies - https://github.com/ernestojimenezruiz/logmap-matcher/releases/tag/logmap-matcher-july-2021
- Sbert - cite: Reimers, N. and Gurevych, I. (2025) Sentence-BERT (Sentence Transformers) Library, Source code [Online]. Available at: https://www.sbert.net/ (Accessed: 8 December 2025)
- LogMap and LogMapBio - cite: * Jiménez-Ruiz, E. (2025) LogMap: An Ontology Alignment and Alignment Repair System, Source code [Online]. Available at: https://github.com/ernestojimenezruiz/logmap-matcher (Accessed: 8 December 2025).