A Scalable Privacy-preserving Data Generation Methodology for Exploratory Analysis.

Area of Science:

• Biomedical Informatics
• Data Science
• Precision Medicine

Background:

• Big data and precision medicine offer potential for understanding complex disorders like cancer and diabetes.
• Biomedical data is often siloed, hindering the identification of relevant datasets for specific analyses.
• Accessing and acquiring biomedical data is effort-intensive, and data may not always yield significant insights.

Purpose of the Study:

• To develop a privacy-preserving method for creating synthetic data.
• To enable the measurement of dataset utility/relevance for biomedical research tasks without direct data access.
• To approximate the utility of additional datasets for specific research needs.

Main Methods:

• Development of a privacy-preserving approach to generate synthetic data.
• Evaluation of the synthetic data generation method using multiple biomedical datasets.
• Assessment of utility approximation for regression and classification tasks.

Main Results:

• The proposed privacy-preserving synthetic data approach provides a first-order approximation of utility.
• The method was evaluated on various biomedical datasets for both regression and classification tasks.
• Demonstrated the potential for synthetic data to guide the selection of relevant datasets.

Conclusions:

• The developed privacy-preserving synthetic data approach can help overcome data silos in biomedical research.
• This method facilitates the assessment of data utility for precision medicine and complex disorder research.
• The approach can be integrated into existing data management systems like REDCap for enhanced data utility analysis.