Quantification of private information leakage from phenotype-genotype data: linking attacks.

Area of Science:

• Genomic privacy and data security
• Bioinformatics and computational biology
• Molecular phenotyping and genetic linkage

Background:

• Traditional genomic privacy focuses on DNA variants, assuming molecular phenotype data (e.g., gene expression) is anonymized.
• Publicly available genotype-phenotype correlations, such as expression quantitative trait loci (eQTLs), can link phenotypes back to genotypes.
• High-dimensional phenotype data increases the accuracy of these statistical linkage attacks, potentially revealing sensitive health information.

Purpose of the Study:

• To develop frameworks for quantifying information leakage from molecular phenotype datasets.
• To enable estimation of data leakage risk before the release of large-scale datasets.
• To present practical methods for performing genotype-phenotype linkage attacks.

Main Methods:

• Development of novel frameworks for assessing information leakage from phenotype data.
• Introduction of a general three-step procedure for executing linkage attacks.
• Implementation of a specific attack leveraging outlier gene expression levels for accuracy.

Main Results:

• Demonstrated frameworks effectively quantify characterizing information leakage from phenotype data.
• The proposed three-step procedure and outlier attack method are shown to be practical and accurate.
• Effectiveness of the outlier attack was evaluated across various data scenarios.

Conclusions:

• Molecular phenotype data is not inherently anonymized and poses genomic privacy risks.
• The developed frameworks provide essential tools for risk assessment of data leakage.
• Practical attack methods highlight the need for robust privacy-preserving techniques in bioinformatics.