Supplementary Figure 10: The accuracy of linking attacks under different scenarios.
From: Quantification of private information leakage from phenotype-genotype data: linking attacks
(a) The distribution of ranks for close relatives (blue) and for random individuals (red) in the linking in 30 HAPMAP CEU trio dataset. Assigned rank is shown in x-axis and frequency is shown on y-axis. (b) The positive predictive value (PPV) versus sensitivity with changing i1,2 threshold for the eQTL selection in (c) where linking accuracy is around 70%, indicated by dashed yellow line. The grey dashed line marks the 95% PPV. The magenta lines show the same plot for random threshold selections. (c) The accuracy of linking attack when the eQTLs are discovered on the training set of 210 individuals and linking is performed on testing set of 211 individuals. The association strength (for eQTL selection) as reported by Matrix eQTL is plotted on x-axis and linking accuracy is plotted on y-axis. (d) The accuracy of linking when the simulated set of 100,211 individuals are used in the genotype dataset, using the eQTLs identified in training sample set in (a).
