We introduce Secure Haplotype Imputation Employing Local Differential privacy (SHIELD), a program for accurately estimating the genotype of target samples at markers that are not directly assayed by array-based genotyping platforms while preserving the privacy of donors to public reference panels. At the core of SHIELD is the Li-Stephens model of genetic recombination, according to which genomic information is comprised of mosaics of ancestral haplotype fragments that coalesce via a Markov random field. We use the standard forward-backward algorithm for inferring the ancestral haplotypes of target genomes, and hence the most likely genotype at unobserved sites, using a reference panel of template haplotypes whose privacy is guaranteed by the randomized response technique from differential privacy.

翻译：我们提出了一种基于局部差分隐私的安全单倍型推断方法（SHIELD），该程序能够在精确估计目标样本在未直接通过阵列基因分型平台检测的标记位点上基因型的同时，保护公共参考面板捐赠者的隐私。SHIELD的核心是Li-Stephens基因重组模型，该模型将基因组信息视为通过马尔可夫随机场溯祖形成的祖先单倍型片段镶嵌体。我们采用标准的前向-后向算法，基于一组模板单倍型参考面板推断目标基因组的祖先单倍型，从而确定未观测位点最可能的基因型。该参考面板的隐私通过差分隐私中的随机响应技术得到保障。