Towards Quantifying and Preventing the Leakage of Genomic Data Using Privacy-Enhancing Technologies

Erman Ayday, Jean Louis Raisaro, Mathias Humbert, Jean‐Pierre HubauxInfoscience 2013

Abstract

Towards Quantifying and Preventing the Leakage of Genomic Data Using Privacy-Enhancing Technologies

Introduction

Your genome is the ultimate identifier—it reveals ancestry, disease predispositions, and physical traits. Unlike a compromised password, you cannot change your DNA. As personalized medicine becomes mainstream, the risk of genomic data leakage grows, potentially leading to genetic discrimination by insurers or employers. This paper addresses a core paradox: medical researchers need access to genomic data to provide life-saving treatments, but storing and sharing this sensitive data in the clear is a massive privacy risk. The authors propose a cryptographic framework that allows for "blind" computation, where a medical center can obtain genetic test results without the storage provider or the center itself ever seeing the patient's full raw DNA sequence.

Proposed Architecture and Methods

The authors propose a decentralized architecture involving three main entities to ensure data security and privacy throughout the testing lifecycle:

  1. Certified Institution (CI): Responsible for sequencing the DNA and encrypting the variants.
  2. Storage and Processing Unit (SPU): An "honest-but-curious" cloud entity that stores encrypted data and performs computations.
  3. Medical Center (MC): Requests specific tests and decrypts only the final result.
The proposed decentralized architecture for privacy-preserving genomic testing, illustrating the flow of data and keys between the Patient, CI, SPU, and Medical Center.

The Cryptographic Toolkit

To enable this architecture, the system utilizes several advanced privacy-enhancing technologies:

  • Paillier Encryption: A form of homomorphic encryption that allows the SPU to perform additions on encrypted data. This is used to calculate weighted disease risks without knowing the underlying genetic values.
  • Proxy Re-encryption: This allows the SPU to transform data encrypted with the patient's key into data that the Medical Center can decrypt, without the SPU ever seeing the plaintext.
  • Asymmetric Entropy: A novel metric introduced to quantify privacy. It accounts for the fact that genomic data isn't random; because of Linkage Disequilibrium (LD), knowing one part of your DNA allows an attacker to make educated guesses about other parts.

Quantifying Privacy as a Depleting Resource

One of the paper's most striking insights is that genomic privacy is not a binary state but a budget that decreases with every medical test. Because Single Nucleotide Polymorphisms (SNPs) are correlated through Linkage Disequilibrium, a single test reveals more than just the specific markers being targeted.

Genomic privacy depletion over successive time-slots (tests), comparing the leakage when markers are treated as independent (no LD) versus when correlations are accounted for (with LD).

As shown in the experimental results, privacy drops significantly faster when an attacker accounts for LD (the "with LD" curve). The model assumes the SPU is "curious" and will attempt to infer sequences; homomorphic encryption effectively neutralizes this threat by ensuring the SPU processes data it cannot read.

Experimental Results

The system was evaluated using real SNP profiles from the 1000 Genomes Project. Implementation on a standard Intel i7-2620M CPU demonstrated that privacy-preserving genomics is computationally feasible for clinical settings:

  • Efficiency: A medical center can decrypt a test result in just 1.4 seconds. Homomorphic operations at the SPU take approximately 100 seconds for 10 variants.
  • Storage: Encrypted genomic data for one patient requires roughly 4.1 GB, a manageable footprint for modern data centers.
  • Security: The system employs a robust 4096-bit security parameter to ensure resilience against brute-force attacks.

Conclusion and Future Work

While the current system is efficient for susceptibility tests involving dozens of SNPs, more complex genomic analyses may require even more efficient homomorphic schemes. Future research will focus on developing techniques to further slow the leakage of genomic data over a lifetime of testing and scaling the system to handle whole-genome sequencing data more holistically. Bridging the gap between these cryptographic tools and clinical practice remains a key goal for the wider adoption of genomic medicine.

Summary updated

Abstract

Towards Quantifying and Preventing the Leakage of Genomic Data Using Privacy-Enhancing Technologies
BibTeX
@inproceedings{ayday2013towards,
  author = {Erman Ayday and Jean Louis Raisaro and Mathias Humbert and Jean‐Pierre Hubaux},
  title = {{Towards Quantifying and Preventing the Leakage of Genomic Data Using Privacy-Enhancing Technologies}},
  booktitle = {Infoscience},
  year = {2013},
}
Figure