bioCADDIE

bioCADDIE provides indexed search and retrieval of publicly accessible biomedical datasets to enable dataset discovery and reuse.

Key Features:

• Information Retrieval System: Implements an advanced IR system tailored to retrieving relevant biomedical datasets from a growing repository based on user queries.
• Utilization of Unstructured Texts: Analyzes unstructured texts such as dataset titles and descriptions to improve retrieval accuracy.
• Medical Named Entity Extraction: Extracts and categorizes biomedical terms from dataset descriptions to enhance search precision.
• Deep Learning-Based Word Embeddings: Employs deep learning-derived word embeddings to expand queries with semantically related terms.
• Re-ranking Strategy: Applies a re-ranking strategy to prioritize datasets according to relevance to expanded queries.

Scientific Applications:

• Dataset discovery: Facilitates locating publicly accessible biomedical datasets relevant to specific research questions.
• Dataset reuse for research and validation: Supports reuse of discovered datasets for hypothesis testing and experimental validation in biomedical studies.

Methodology:

Analyzes unstructured text (titles, descriptions), applies medical named entity extraction, uses deep learning-based word embeddings for query expansion, employs re-ranking of search results, and was evaluated in the bioCADDIE Dataset Retrieval Challenge benchmarked against 11 baseline systems using inference Average Precision and inference normalized Discounted Cumulative Gain.