Extraction and Curation of Glycan Biomarker Data from Biomedical Literature Using Large Language Models

Automated extraction and curation of glycan biomarker data from biomedical literature are crucial for synthesizing discrete findings into generalized glycan-disease associations. This thesis presents a scalable named entity recognition (NER) and relation extraction (RE) pipeline utilizing state-of-the-art large language models (LLMs), specifically GPT-based models, to systematically identify glycans, glycoproteins, glycoenzymes, diseases, and experimental contexts from a corpus of 4,595 glycan biomarker-related PubMed abstracts, yielding ~300 distinct glycan biomarker entries. The pipeline accurately converts complex biomarker relationships described in unstructured natural language texts into structured data adhering to a custom-designed JSON schema, enabling efficient integration into downstream analytical processes such as clustering and embedding analyses. Validation against a manually curated dataset of 500 abstracts confirms the approach’s high precision and reliability, implying a potential replacement for manual annotation efforts traditionally required for glycoinformatics research. Distinct from previous text mining efforts which relied primarily on domain-specific NLP approaches, this method leverages the contextual understanding and flexibility of general-purpose LLMs. The pipeline offers unique advantages, including strict schema compliance and efficient parallel processing, hence facilitating large-scale biomedical literature mining without requiring significant local computational resources. By enabling comprehensive curation of glycan-diseases relationships, this study provides a robust foundation for hypothesis generation and glycan-associated disease mechanisms.

Author

• Au Yeung, Cyrus Chun Hong

Language

• en

Keyword

• Large Language Models
• Glycosylation
• Natural Language Processing
• Biomarkers

Date created

• 2025

Type of Work

• Master's Thesis

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Bioinformatics and Molecular Biochemistry

Degree

• M.S.

Advisor

• Mazumder, Raja
• Kahsay, Robel

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/79407z20r