Protecting Genomic Data Privacy with Probabilistic Modeling.

Area of Science:

• Genomics
• Biomedical Research
• Data Privacy

Background:

• Sequencing technologies generate large-scale genomic data, raising privacy concerns for aggregate data.
• Existing privacy methods like differential privacy can significantly reduce data accuracy.
• There is a need for privacy-preserving methods that maintain the utility of aggregate genomic data.

Purpose of the Study:

• To develop an alternative approach for privacy-preserving aggregate genomic data sharing.
• To ensure privacy without the high accuracy costs associated with differential privacy.
• To apply statistical disclosure control principles to aggregate genomic data.

Main Methods:

• Utilizing concepts from statistical disclosure control, specifically disclosure risk.
• Combining minimal data perturbation with Bayesian statistics.
• Employing Markov Chain Monte Carlo (MCMC) techniques for privacy preservation.
• Testing the proposed method on a Genome-Wide Association Study (GWAS) dataset.

Main Results:

• Demonstrated a novel technique for privacy-preserving aggregate genomic data sharing.
• Achieved strong privacy guarantees with minimal impact on data accuracy.
• Validated the method's practical utility using a real-world GWAS dataset.

Conclusions:

• The proposed method offers a viable alternative to differential privacy for secure genomic data sharing.
• It effectively balances privacy protection with analytical accuracy in aggregate genomic data.
• This approach enhances the responsible sharing of sensitive genomic information.