Assessing transcriptomic reidentification risks using discriminative sequence models.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene expression quantitative trait loci (eQTLs) link genetic variation to gene expression.
• Sharing omics data raises privacy concerns due to potential re-identification of individuals.
• Previous methods for assessing linking attack risks were limited by restrictive assumptions.

Purpose of the Study:

• To develop a novel framework for predicting genotypes from gene expression data.
• To enhance the power and accuracy of linking attacks on omics datasets.
• To provide a unified approach for assessing privacy risks in diverse omics data.

Main Methods:

• Introduced the Discriminative Sequence Model (DSM), a probabilistic framework.
• Modeled the joint distribution of eQTLs within genomic regions.
• Incorporated calibration for linkage disequilibrium and redundant signals.

Main Results:

• DSM significantly improves linking attack accuracy compared to existing methods.
• Demonstrated enhanced linking power across various attack scenarios and datasets.
• Identified substantial additional privacy risks overlooked by prior studies.

Conclusions:

• DSM provides a more comprehensive assessment of privacy risks associated with gene expression data.
• The framework is applicable to diverse omics datasets beyond transcriptomics.
• Highlights the need for robust privacy-preserving strategies in omics data sharing.