Preserving Privacy for Distributed Genome-Wide Analysis Against Identity Tracing Attacks

Abstract

Genome-wide analysis has demonstrated both health and social benefits. However, large scale sharing of such data may reveal sensitive information about individuals. One of the emerging challenges is identity tracing attack that exploits correlations among genomic data to reveal the identity of DNA samples. In this paper, we first demonstrate that the adversary can narrow down the sample's identity by detecting his/her genetic relatives and quantify such privacy threat by employing a Shannon entropy-based measurement. For example, we exemplify that when the dataset size reaches 30% of the population, for any target from that population, the uncertainty of the target's identity is reduced to merely 2.3 bits of entropy (i.e., the identity is pinned down within 5 people). Direct application of existing approaches such as differential privacy (DP), secure multiparty computation (MPC) and homomorphic encryption (HE) may not be applicable to this challenge in genome-wide analysis because of the compromise on utility (i.e., accuracy or efficiency). Towards addressing this challenge, this paper proposes a framework named υFRAG to facilitate privacy-preserving data sharing and computation in genome-wide analysis. υFRAG mitigates privacy risks by using a vertical fragmentation to disrupt the genetic architecture on which the adversary relies for identity tracing without sacrificing the capability of genome-wide analysis. We theoretically prove that it preserves the correctness of the primitive functionalities and algorithms ranging from basic summary statistics to advanced neural networks. Our experiments demonstrate that υFRAG outperforms secure multiparty computation (MPC) and homomorphic encryption (HE) protocols, with a speedup of more than 221x for training neural networks, and also traditional non-private algorithms and a state-of-the-art noise-based differential privacy (DP) solution in most settings.

Genomic privacy preservation in genome-wide association studies: taxonomy, limitations, challenges, and vision

“…Yet, despite a wide range of existing techniques for privacy-preserving GWAS analysis, including differential privacy, deep learning frameworks, and cryptographic technologies such as homomorphic encryption or secure multi-party computation, we are still so far from a generic framework, which may be applicable to genome-wide analysis, due to some inherent privacy and security risks [ 50 ]. Our review analysis revealed that most of the proposed privacy-preserving methods are limited to some statistics [ 20 ], as they do not cover all the GWAS tests [ 82 ].…”

“…Indeed, only a few studies can provide a correctness verification mechanism for the computation of the federated learning model [ 66 ]. Even worse is that only a limited number of studies have considered these threats in the GWAS context [ 50 ]. So far, verifying the integrity of the federated learning model and the correctness of the outcomes in an environment handled by intruders and attackers is not a trivial task [ 85 ], as the server does not have access to the dataset [ 59 ].…”

“…Differential privacy is another approach that has gathered much more interest over the proposed methods to guarantee the privacy preservation in the past years [ 50 ]. By adding noise to the data or model parameters, differential privacy can ensure privacy protection against attacks [ 51 ].…”

“…By adding noise to the data or model parameters, differential privacy can ensure privacy protection against attacks [ 51 ]. Several studies have explored the differential private regardless of the privacy protection of genetic GWAS data [ 50 ]. Dervishi et al .…”

Better Safe Than Sorry: Constructing Byzantine-Robust Federated Learning with Synthesized Trust

“…Federated learning (FL) [1][2][3][4][5][6] is a cooperative machine learning technique that makes use of several clients (users) and a centralized server. Each client keeps track of its own private dataset, and only transfers its local parameter updates (gradient or weight) during iterative processes.…”