Addressing the Threats of Inference Attacks on Traits and Genotypes from Individual Genomic Data

摘要

The decreasing cost of DNA-sequencing empowers high availability of genetic-oriented services, which further promote growing number of genomes and traits of individuals being accessible online. Notoriously, these data are sensitive and may further lead to more sensitive data leakage. In this paper, we formulate the trait and genotype inference problem and develop an efficient inference method based on factor graph and belief propagation. An adversary then can infer the potential traits and genotypes of the victims whose portions of data are observed, depending on trait/SNP associations available from GWAS catalog. To protect against such inference attacks, we detail privacy and utility metrics then propose a genomic data-sanitization method that can effectively tradeoff genomic data openness and privacy.