PROVGEN: A Privacy-Preserving Approach for Outcome Validation in Genomic Research.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic dataset sharing is limited by privacy concerns, hindering research reproducibility and validation.
• Reproducibility and validation are crucial for identifying computational errors in genomic research.

Purpose of the Study:

• Introduce PROVGEN, a privacy-preserving method for sharing genomic datasets.
• Facilitate reproducibility and outcome validation in genome-wide association studies (GWAS).

Main Methods:

• Encode genomic data into binary space.
• Apply a two-stage process: differentially private data generation using an XOR-based mechanism and data utility restoration via optimal transport-based MAF adjustment.
• Convert processed binary data back to genomic representation for publication.

Main Results:

• PROVGEN outperforms existing methods in detecting GWAS outcome errors.
• Demonstrates superior data fidelity preservation and resistance to membership inference attacks (MIAs).
• Evaluated on three real-world genomic datasets against local differential privacy and synthesis-based methods.

Conclusions:

• PROVGEN encourages sharing of differentially private genomic datasets.
• Maintains high data quality essential for research reproducibility.
• Addresses privacy concerns while supporting robust scientific validation.