Semantic Harmonization of Alzheimer's Disease Datasets Using AD-Mapper

Area of Science:

• Biomedical Informatics
• Computational Biology
• Neuroscience Data Science

Background:

• Semantic harmonization of Alzheimer's disease (AD) cohort studies remains a challenge.
• Existing methods lack automated solutions for cross-cohort data integration.
• Data-driven analysis in AD research relies heavily on harmonized cohort data.

Purpose of the Study:

• To accelerate the semantic harmonization of AD cohort studies.
• To develop an automated tool for improving cross-cohort analysis.
• To enhance data consistency and comparability across diverse AD datasets.

Main Methods:

• Created a common data model (CDM) by cross-mapping data from 20 cohorts, three CDMs, and ontology terms.
• Fine-tuned a BioBERT model using the developed CDM.
• Evaluated the model's performance on three unseen cohorts against a string-matching baseline.

Main Results:

• Introduced AD-Mapper, an interface for automatic harmonization of AD cohort studies.
• AD-Mapper demonstrated superior performance compared to a string-matching baseline on unseen cohorts.
• Showcased a comprehensive CDM containing 1218 unique variables.

Conclusions:

• AD-Mapper effectively leverages semantic similarities in naming conventions for improved harmonization.
• The tool offers a robust solution for the long-standing challenge of AD cohort data harmonization.
• Facilitates more reliable and efficient data-driven research in Alzheimer's disease.