Privacy considerations for sharing genomics data

Marie Oestreich, Dingfan Chen, Joachim L. Schultze, Mario Fritz, Matthias BeckerPubMed 2021 · DOI

TL;DR

Sharing genomic data is crucial for advancing personalized medicine and research, but it poses significant privacy risks due to the potential for re-identification. This review explores these risks across different genomic data types (sequence, transcriptomic, epigenomic), discusses motivations for sharing, and surveys existing and emerging privacy-preserving techniques and legal frameworks.

What Problem Does This Paper Address?

The paper addresses the growing challenge of balancing the immense benefits of sharing human genomic data for research and medical advancement with the critical need to protect individual privacy. As genomic data generation becomes more accessible and diverse, so do the methods for re-identifying individuals, leading to potential social stigma, discrimination in employment and insurance, and even impacting relatives. The authors aim to provide a comprehensive overview of these privacy concerns, the motivations driving data sharing, and the current landscape of solutions and legal protections.

Methods

This paper is a review article, so it doesn't introduce new experimental methods. Instead, it synthesizes existing research and knowledge. The authors categorize genomic data into three main types:

  • Sequence Data: Information about the specific base sequence of an individual's genome (e.g., Whole Genome Sequencing, Whole Exome Sequencing, SNPs).
  • Transcriptomic Data: Quantitative data on RNA transcript abundance, influenced by both genomic sequences and environmental factors.
  • Epigenomic Data: Information on heritable alterations affecting gene expression that are not changes to the DNA sequence itself (e.g., DNA methylation, histone modifications).

They then discuss various privacy threats and established and emerging data sharing techniques, including:

  • Established Methods: Free access, controlled access (with and without download), Safe Harbor, data sanitization, fully homomorphic encryption, secure multi-party computation, and differential privacy.
  • Emerging/Research Methods: Federated learning, swarm learning, and synthetic data generation (using GANs, RBMs, VAEs).

The review also examines relevant legal frameworks and limitations.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Re-identification is the Core Threat: The fundamental privacy concern across all genomic data types is the potential for re-identification, which can have severe social and personal consequences.
  • Interconnectedness of Data Types: Privacy risks can cascade across different genomic data types. For instance, transcriptomic or epigenomic data can sometimes be used to infer underlying sequence information, amplifying re-identification risks.
  • SNP "Barcoding" is Powerful: Even a small number of Single Nucleotide Polymorphisms (SNPs) can act as a unique "fingerprint," allowing individuals to be identified by cross-referencing with public databases.
  • Metadata is a Major Vulnerability: Personal information contained in metadata associated with genomic datasets can significantly increase re-identification risk, even without direct access to the sequence data itself.
  • Motivation for Sharing is Growing: Beyond traditional research needs, the rise of machine learning, personalized medicine, and the desire for reproducibility are driving increased demand for genomic data sharing.
  • Legal Frameworks are Evolving but Ambiguous: While laws like GINA (US) and GDPR (EU) aim to protect against genetic discrimination, their definitions of "genetic data" can be ambiguous, leaving gaps in protection for newer data types.
  • Privacy-Presuring Techniques Have Trade-offs: Methods like differential privacy offer strong guarantees but often come at the cost of reduced data utility, while others like homomorphic encryption are computationally intensive.
  • Future is Collaborative and Proactive: Addressing genomic data privacy requires ongoing dialogue between researchers, lawmakers, and the public, alongside the development of more robust, integrated privacy-preserving technologies.

Key Results

The paper is a review, so it doesn't present new experimental results. However, it highlights key findings from the literature:

  • SNP Fingerprinting: As few as 30-80 independent SNPs can be sufficient for unequivocal individual identification.
  • Transcriptomic Data Risk: Studies have shown that it's possible to infer genotypes at eQTL positions from publicly available gene expression data, enabling SNP-fingerprinting.
  • Epigenomic Data Risk: Membership inference attacks can be used on DNA methylation data to infer an individual's presence in a dataset, and this data can also potentially reveal smoking status or alcohol consumption.
  • Legal Landscape: The paper outlines various laws (HIPAA, GINA, GDPR, etc.) and their limitations, noting that definitions of genetic data are often not comprehensive enough for modern genomic data types.

Limitations and Future Work

  • Evolving Threat Landscape: The authors emphasize that privacy risks are not static; as technology advances, new attack vectors will emerge, requiring continuous reassessment.
  • Utility vs. Privacy Trade-off: Many privacy-preserving techniques still face challenges in balancing robust privacy guarantees with sufficient data utility for meaningful analysis.
  • Need for Standardized Metadata: A synchronized metadata vocabulary is needed to avoid varying levels of information being revealed by different data descriptors.
  • Legal Ambiguity: Current laws often have ambiguous wording regarding different types of genomic data, necessitating clearer definitions and broader applicability.
  • Future Research Directions: The paper calls for further research into:
  • Improving the utility of generative models for synthetic data while minimizing information leakage.
  • Developing privacy-preserving techniques for genomics data beyond just sequences.
  • Fostering active communication between researchers and lawmakers to create up-to-date, non-discriminatory laws.
  • Enhancing collaboration between data security/privacy researchers and life scientists to develop tailored genomic data sharing techniques.
  • Increased necessity for bioinformaticians and computational biologists to communicate genomics needs to computer scientists for easier, secure data sharing.

Summary updated

Abstract

An increasing amount of attention has been geared towards understanding the privacy risks that arise from sharing genomic data of human origin. Most of these efforts have focused on issues in the context of genomic sequence data, but the popularity of techniques for collecting other types of genome-related data has prompted researchers to investigate privacy concerns in a broader genomic context. In this review, we give an overview of different types of genome-associated data, their individual ways of revealing sensitive information, the motivation to share them as well as established and upcoming methods to minimize information leakage. We further discuss the concise threats that are being posed, who is at risk, and how the risk level compares to potential benefits, all while addressing the topic in the context of modern technology, methodology, and information sharing culture. Additionally, we will discuss the current legal situation regarding the sharing of genomic data in a selection of countries, evaluating the scope of their applicability as well as their limitations. We will finalize this review by evaluating the development that is required in the scientific field in the near future in order to improve and develop privacy-preserving data sharing techniques for the genomic context.
BibTeX
@inproceedings{oestreich2021privacy,
  author = {Marie Oestreich and Dingfan Chen and Joachim L. Schultze and Mario Fritz and Matthias Becker},
  title = {{Privacy considerations for sharing genomics data}},
  booktitle = {PubMed},
  year = {2021},
  doi = {10.17179/excli2021-4002},
}
Figure