Needles in the haystack: identifying individuals present in pooled genomic data

Abstract

Recent publications have described and applied a novel metric that quantifies the genetic distance of an individual with respect to two population samples, and have suggested that the metric makes it possible to infer the presence of an individual of known genotype in a sample for which only the marginal allele frequencies are known. However, the assumptions, limitations, and utility of this metric remained incompletely characterized. Here we present empirical tests of the method using publicly accessible genotypes, as well as analytical investigations of the method's strengths and limitations. The results reveal that the null distribution is sensitive to the underlying assumptions, making it difficult to accurately calibrate thresholds for classifying an individual as a member of the population samples. As a result, the false-positive rates obtained in practice are considerably higher than previously believed. However, despite the metric's inadequacies for identifying the presence of an individual in a sample, our results suggest potential avenues for future research on tuning this method to problems of ancestry inference or disease prediction. By revealing both the strengths and limitations of the proposed method, we hope to elucidate situations in which this distance metric may be used in an appropriate manner. We also discuss the implications of our findings in forensics applications and in the protection of GWAS participant privacy.

Figure 1. Comparison of T distributions.

Comparison of T distributions for true positive and null samples versus putative null distribution, starting with 481,382 SNPs in (A,B) and 50,000 SNPs in (C,D). In all plots, true positive (1042 CGEMS controls) is shown as a solid green curve, true positive (1045 CGEMS cases) is shown as a solid red curve, and the putative null is given as a thin grey curve. The dark and light grey regions represent the areas for which the null hypothesis would be accepted at and , respectively. In plots (A,C), CGEMS test samples in neither nor (100 CGEMS cases and 100 CGEMS controls) are given by a heavy black curve. The CGEMS case and CGEMS control distributions within this group are shown as dashed red and green lines, respectively. In plots (B,D), distributions are given for HapMap CEPHs (cyan) and YRIs (blue). Vertical lines mark the 0.05 and 0.95 quantiles of the negative CGEMS samples (black), HapMap CEPHs (cyan), and HapMap YRIs (blue).

Figure 2. Distribution of T.

Distributions of T for out-of-group samples who are related (red line) and unrelated (blue line) to individuals in G for HapMap YRI (A) and HapMap CEPH (B) populations. (C) and (D) show the same distributions as (A) and (B) respectively, with the addition (green line) of individuals who are in G and unrelated to F (i.e., true positives). Dashed black lines indicate the T significance thresholds of ±1.64 at nominal .

Figure 3. Positive predictive value (PPV) as a function of prevalence and specificity given 99% sensitivity.

In (A), PPV is shown on the y axis and color corresponds to specificity. The black curve depicts the 87% sensitivity line—the best sensitivity obtained in the empirical tests in Table 2. In (B), PPV is shown by color, and the y axis corresponds to specificity.