One Size Doesn't Fit All: Measuring Individual Privacy in Aggregate Genomic Data

Area of Science:

• Genomics
• Bioinformatics
• Privacy-preserving data analysis

Background:

• Aggregate genomic data, such as minor allele frequencies (MAFs), can inadvertently reveal sensitive individual information.
• Existing privacy measures often focus on aggregate privacy loss, not individual participant risk.
• There is a need for methods that provide individual-level privacy guarantees in genomic studies.

Purpose of the Study:

• To introduce PrivMAF, a novel model-based measure for provable individual privacy guarantees in aggregate genomic data.
• To quantify privacy gains achieved through data perturbation techniques like noise addition and binning.
• To assess the practical privacy risks in genome-wide association studies (GWAS) using real-world data.

Main Methods:

• Development of a model-based measure (PrivMAF) to assess individual privacy loss for each participant.
• Application of PrivMAF to genotype data from the Wellcome Trust Case Control Consortium.
• Quantification of privacy gains from data perturbation methods (noise addition, binning) and their impact on data utility.
• Comparison of privacy gains against stricter privacy notions like differential privacy.

Main Results:

• PrivMAF provides individual privacy measures, enabling the assessment of worst-case privacy loss within a study.
• Data perturbation techniques (noise addition and binning) demonstrably yield significant privacy gains.
• These privacy gains can be achieved with minimal perturbation, thus maximizing data utility.
• Individual privacy risks associated with releasing MAFs can vary substantially among participants in a study.

Conclusions:

• PrivMAF offers a robust framework for evaluating and ensuring individual privacy in aggregate genomic datasets.
• Perturbation methods are effective in enhancing privacy without critically compromising data utility.
• The study highlights the importance of individual-level privacy assessment in genomic research, moving beyond aggregate measures.