Genomic Data Sharing under Dependent Local Differential Privacy.

Area of Science:

• Genomics
• Computer Science
• Bioinformatics

Background:

• Genomic data sharing accelerates research and personalized medicine.
• Existing local differential privacy (LDP) methods are insufficient for correlated genomic data.
• Protecting sensitive genomic information during sharing is crucial.

Purpose of the Study:

• To propose a novel privacy-preserving mechanism for sharing correlated genomic data.
• To address the limitations of standard LDP in the context of genomic data.
• To enhance both data privacy and utility for genomic research.

Main Methods:

• Introduction of (ϵ, T)-dependent local differential privacy (LDP).
• Development of a genomic data sharing mechanism that accounts for data correlations.
• Pre-elimination of statistically unlikely data values and adjustment of probability distributions.
• Implementation of a greedy algorithm for optimizing data processing order.

Main Results:

• The proposed mechanism effectively prevents attackers from inferring private genomic data using correlations.
• Adjusting probability distributions improves the utility of shared data for collectors.
• Evaluations on a real-life genomic dataset demonstrate superior performance over the randomized response mechanism.
• The greedy algorithm successfully maximizes the utility of shared data.

Conclusions:

• The (ϵ, T)-dependent LDP framework offers a robust solution for privacy-preserving genomic data sharing.
• The proposed mechanism balances privacy protection with data utility, advancing personalized genomic medicine.
• This approach represents a significant improvement over existing LDP techniques for genomic applications.