On Utility and Privacy in Synthetic Genomic Data

Abstract

The availability of genomic data is essential to progress in biomedical research, personalized medicine, etc. However, its extreme sensitivity makes it problematic, if not outright impossible, to publish or share it. As a result, several initiatives have been launched to experiment with synthetic genomic data, e.g., using generative models to learn the underlying distribution of the real data and generate artificial datasets that preserve its salient characteristics without exposing it.This paper provides the first evaluation of both utility and privacy protection of six state-of-the-art models for generating synthetic genomic data. We assess the performance of the synthetic data on several common tasks, such as allele population statistics and linkage disequilibrium. We then measure privacy through the lens of membership inference attacks, i.e., inferring whether a record was part of the training data. Our experiments show that no single approach to generate synthetic genomic data yields both high utility and strong privacy across the board. Also, the size and nature of the training dataset matter. Moreover, while some combinations of datasets and models produce synthetic data with distributions close to the real data, there often are target data points that are vulnerable to membership inference. Looking forward, our techniques can be used by practitioners to assess the risks of deploying synthetic genomic data in the wild and serve as a benchmark for future work.

Future-proofing genomic data and consent management: a comprehensive review of technology innovations

“…A recent study by Oprisanu et al. [ 103 ] compared the above synthetic data methods for utility and privacy. They showed that recombination-based methods have high utility but low privacy, while RBMs offer a trade-off.…”

Synthetic data as an enabler for machine learning applications in medicine

“…Outliers are a particularly clear example of this trade-off because of their fundamental difficulty to be statistically captured based on their uniquely identifying features. If the utility is based on learning from outliers, then a useful and private SDG will be challenging, see ( Oprisanu et al., 2022 ) for a demonstration in SDG of genomic data. Therefore, an ideal private synthetic dataset is created by solving the privacy-utility trade-off (see Figure 1 ) optimized to the needs of all the stakeholders.…”

Assessing Privacy Vulnerabilities in Genetic Data Sets: Scoping Review

“…However, it has become evident that even this limited amount of information can be exploited for privacy attacks, and few queries to genomic beacons can suffice to determine whether individuals (whose genome is known) are present in a study cohort [ 20 - 23 ]. Similarly, proposals for encryption and differential privacy approaches [ 24 , 25 ] have often been countered by demonstrations of attacks [ 26 - 28 ], and even synthetic genetic data may not fully protect the study participants from privacy attacks [ 29 ] (refer to the study by Mittos et al [ 30 ] for a review of privacy-enhancing technologies). Thus, even a substantial reduction in information content can often not completely eliminate all privacy risks of genetic data [ 31 ].…”