Private information leakage from single-cell count matrices.

Area of Science:

• Genomics
• Bioinformatics
• Data Privacy

Background:

• Growing availability of large-scale human single-cell datasets enhances biological understanding.
• Data accessibility raises significant privacy concerns for individuals.
• Previous privacy studies focused on bulk gene expression data due to noise in single-cell data.

Purpose of the Study:

• To demonstrate privacy vulnerabilities in single-cell gene expression datasets.
• To develop methods for genotype prediction and linking genotype to phenotype.
• To assess the risk of linking attacks using external genetic information.

Main Methods:

• Linking attacks using publicly available expression quantitative trait loci (eQTLs) data.
• Development of a genotype prediction method.
• Genotype-phenotype linking without relying on eQTLs.
• Cross-study analysis of variant exploitation.

Main Results:

• Individuals in single-cell datasets are vulnerable to linking attacks.
• Sensitive phenotypic information can be inferred using eQTLs.
• A novel method enables genotype prediction and phenotype linking without eQTLs.
• Genetic variants from one study can compromise privacy in another.

Conclusions:

• Single-cell gene expression data poses significant privacy risks.
• Linking attacks can exploit publicly available genetic information.
• New methods are needed to protect individual privacy in large-scale genomic datasets.