BDI-Kit: An AI-powered toolkit for biomedical data harmonization

Figures

• 

  Figure 1. BDI-Kit architecture Given source and target datasets (or common data model), users can construct pipelines that orchestrate schema- and value-matching primitives with human-in-the-loop assessment and refinement of their outputs. Users can interact with BDI-Kit through a Python API or by using AI assistants to generate a harmonized dataset along with a reusable harmonization specification. Contributors can extend BDI-Kit by adding new primitives and support for additional data models.

• 

  Figure 2. Example of BDI-Kit code snippet in Python The example shows how to use the functions match_schema(), match_values(), and merge_mappings() to create a harmonization specification and generate a new harmonized dataset.

• 

  Figure 3. Example of BDI-Kit integration with an AI assistant or agent On the Python side (top), the MCP server is executed, while in the AI assistant or agent side (bottom), users can issue queries and receive answers in natural language.

• 

  Figure 4. A snippet from the harmonization specification The example shows the harmonization specification for the attributes Histologic_Grade_FIGO and Path_Stage_Primary_Tumor.

• 

  Figure 5. Results of calling match_schema() and match_values() functions (A) Attributes and their corresponding values were correctly aligned despite spelling variations. (B) Precise matching of identifiers. (C) Accurate attribute matching even when values contain typos.

• 

  Figure 6. Sample results of attribute matches with their explanations (A) Sample of top-ranked schema matches retrieved for Assay and Organism attributes. (B) Evaluation of these matches, highlighting accepted and rejected matches.

• 

  Figure 7. BDI-Kit accessed directly from an AI agent (A) Initial schema matches are generated in response to the user’s request. (B) Alternative matches for the Age attribute obtained using the rank_schema_matches() function. (C) Selection of the method for value matching in numeric attributes. (D) Example of the transformation applied to convert Age values.