Department of Science, Yale University, New Haven, United States of America
Research Article
Secure Haplotype Imputation Employing Local Differential Privacy
Author(s): Marc Harary*
Recent literature has highlighted the security risks to human research subjects forfeiting sensitive genomic data
to public reference panels. Such breaches to subject privacy may occur even in largescale biomedical analyses like
genotype imputation, a preliminary stage of many clinical studies. To this end, we introduce Secure Haplotype
Imputation Employing Local Differential Privacy (SHIELD). As a server-side pipeline, it combines the differentially
private randomized response mechanism and the standard forward-backward algorithm to compute a Markov
random field over incomplete genomic datasets submitted by client researchers. Critically, we show that SHIELD
achieves modern imputation accuracy well within typical privacy budgets, providing mathematically provable privacy
guarantees to reference panel donors without sacrificing client utility. We conclude that dev.. View more»