Unveiling the risks: protecting privacy in single-cell genomics data

Nephrology Dialysis Transplantation, vol. 40, pp. 1077–1080

Abstract

In recent years, the fields of genomics and single-cell RNA sequencing (scRNA-seq) have emerged as groundbreaking technologies that enable scientists to dissect complex biological mechanisms, drug responses and individual genetic variability with unprecedented resolution. From studying tumour heterogeneity to mapping neural circuits, the insights offered by scRNA-seq have transformed biomedical research, enabling discoveries that would have been unthinkable just a decade ago. However, with these advances come new challenges—among the most pressing is the issue of privacy breaches. The high identifiability potential of genetic data, combined with its sensitive content with regards to health (e.g. susceptibility to diseases) and physical traits, has raised public concern that genetic data shared or published in the context of research or healthcare could be misused [1, 2]. The recent study, Private Information Leakage from Single-Cell Count Matrices, by Walker et al. published in Cell (2024), highlights these concerns by revealing that scRNA-seq count matrices from individuals, similar to bulk RNA sequencing that are usually shared, can, when linked with phenotype data, harbour the risk of re-identification of individuals (Fig. 1) [3]. As highlighted by the authors, even when anonymized using publicly available data on expression quantitative trait loci (eQTL) to match genotype information from scRNA-seq data to a phenotype, such as a specific disease, these datasets still have the potential to inadvertently reveal sensitive information about individuals. To address this, the authors used the OneK1K cohort with 14 annotated immune cell types [4] and companion systemic lupus erythematosus (SLE) [5] studies to demonstrate that individuals (genotype) can be effectively linked to their phenotype, even without relying on tissue- or cell type–specific eQTLs. This study underscores the privacy risks associated with sharing scRNA-seq data, as variants from one study can be used to extract private information about individuals in another, revealing the vulnerabilities in sharing such data [3]. Schematic illustration of the potential privacy risk associated with single-cell data release and genotype prediction. (Top) scRNA-seq data from multiple individuals (indicated as ‘individual 1’, ‘individual 2’, ‘individual 3’ etc.) is collected, capturing gene expression profiles for thousands of individual cells. Using gene expression data from the single-cell count matrix, genotypes for specific variants can be predicted by correlating gene expression with known eQTL from publicly available databases such as bulk tissue eQTLs, cell type–specific eQTLs or AAF-based variants. (Left) The genotype prediction process results in noisy or uncertain genotypes for individuals, depicted as ‘?’ for uncertain variant calls. The predicted noisy genotypes can be linked to a pre-existing reference genome database that contains the true genotypes. (Right) Once the genotypes are successfully linked between the predicted (noisy) genotypes from the single-cell dataset and the genotypes in the reference genome database, individual identities can be re-identified, along with their associated phenotypes (e.g. disease status or other sensitive information). Such scRNA-seq data may even enable the detection of biological information or profiles that can be traced back to a group of individuals, such as those classified as having the same disease. Figure modified from Walker et al. [3] and generated with BioRender. AAF: alternate allele frequency; ID: identity; Ind: individual; Pheno: phenotype. scRNA-seq has revolutionized molecular biology and immunology by allowing researchers to explore gene expression patterns at the level of individual cells. Unlike bulk RNA sequencing, which provides an average gene expression profile for a population of cells, scRNA-seq captures the variability of individual cells, providing a far more nuanced view of biological processes. This ability to ‘zoom in’ on the molecular workings of individual cells has proven invaluable in a variety of research areas, including cancer biology, developmental biology, nephrology and immunology, while the sparsity of single-cell data leads to new challenges that need to be addressed. While reports have addressed the challenges of protecting information privacy [6] in bulk RNA sequencing, this has not yet been considered in the single-cell genomics field. As scRNA-seq becomes increasingly popular, vast datasets containing gene expression profiles from thousands of cells are being generated, shared among researchers and made accessible to the public. The sharing of such datasets is crucial for scientific progress, enabling other researchers to validate findings and build on existing knowledge. But as Walker et al. [3] indicate, these anonymized datasets can inadvertently reveal sensitive information about individuals, even when the data is anonymized using publicly available data on eQTL to match the genotype information from scRNA-seq data to the phenotype. For example, consider a large cohort study collecting eQTL data at the population level, which links single nucleotide polymorphisms (SNPs) to gene expression levels, and an scRNA-seq dataset from individual patients. Potential privacy breaches could occur if a publicly accessible eQTL dataset contains summary statistics associating specific SNPs with gene expression changes across tissues. Simultaneously, a disease-specific scRNA-seq dataset (e.g. from patients with cardiomyopathy) is privately shared among collaborators. An attacker could correlate these gene expression profiles from the single-cell dataset with those predicted by SNPs in the eQTL dataset, potentially inferring the presence of specific genetic variants in individual patients and uncovering sensitive genetic information, including an individual's predisposition to diseases like cardiovascular or kidney disorders. At first glance, it might seem counterintuitive that anonymized gene expression databases could lead to privacy violations. After all, these datasets do not contain traditional identifiers such as names, addresses, date of birth or social security numbers. Yet, as Walker et al. [3] and others [6, 7] reveal, the information encoded in single-cell count matrices can, in fact, be used in theory to re-identify individuals or infer private attributes if linked with datasets from other sources. The risk arises from the fact that gene expression data are not just a snapshot of a cell's function, they also reflect the underlying genetics and physiological state of the individual from whom the cells were derived. By analysing these patterns, it is theoretically possible to cross-reference the data with existing genetic databases, identify individuals and even uncover information about their health status [6]. For instance, the presence of specific mutations or gene expression signatures can suggest an individual's susceptibility to certain diseases or reveal their carrier status for genetic disorders. While datasets on very rare diseases with only a few affected patients could theoretically enable identification of individual patients based on phenotypic information via eQTLs combined with details like location (e.g. country, hospital) disclosed in publications, re-identification in most other cases typically requires genomic information from the individual being present in a public database, such as through SNP-based matching [6]. The potential for privacy leakage from scRNA-seq data has some implications for biomedical research. The principle of open data sharing, which facilitates collaboration, accelerates discovery and ensures the reproducibility of results is important in science. In genomics and transcriptomics, the sharing of large datasets is essential for identifying common patterns, testing new hypotheses and developing new therapeutic approaches. However, the revelation that scRNA-seq and other genetic datasets may not be as anonymous as previously thought challenges this open-data paradigm. As with other genomic studies, in scRNA-seq studies patients must be informed and give consent that their data are shared and that some or all of their genomic information will be made publicly accessible. Beyond the scientific community, there are also legal and ethical concerns to consider. Regulations such as the Health Insurance Portability and Accountability Act (HIPAA) in the USA [8, 9] and the General Data Protection Regulation (GDPR) in Europe [10] impose strict rules on how personal health information can be collected, stored and shared. However, these do not account for the possibility that an individual's genetic data when linked with other data could serve as a unique identifier. Privacy laws might need to be updated to reflect the new risks posed by genomic and single-cell data. Basic ethics principles apply to all research [11], including respecting the autonomy of research subjects by obtaining fully informed consent, avoiding harm to participants, balancing potential benefits with potential risks, ensuring fair treatment of participants, ensuring privacy and confidentiality and maximizing scientific validity while maintaining research integrity and avoiding conflicts of interest. The premise of anonymity in genomic data has been fundamental to ethical considerations in genomics research. However, the reality that this may no longer be possible to ensure poses an important threat for genomics research. This threat not only relates to the identification of an individual's medical information, but may be used to profile individuals and groups/communities—socially and biologically—whether accurately or inaccurately. This can lead to serious consequences, including individual and societal risks, as well as undermining public trust in research and research institutions. These risks are especially important to consider since genomics research often does not directly benefit the individual participant, instead contributing to broader knowledge and future benefits for others. Consequently, the risk:benefit ratio for an individual may become too high if privacy cannot be guaranteed. As these concerns are increasingly becoming realities, researchers, research ethics committees and the public must weigh issues of informed consent, privacy and confidentiality, benefits and harms and the intrinsic value of research itself. Solutions to these challenges are not straightforward; once genomic and other datasets are shared or made publicly accessible, they become vulnerable to misuse and malicious intent. Walker et al. and others propose several strategies to mitigate the risk of private information leakage from genomics and scRNA-seq data (Fig. 1) [3, 12]. One approach is to develop more sophisticated anonymization techniques that take into account the unique properties of single-cell data. For instance, differential privacy—a technique that adds noise to the data in a way that preserves overall trends while masking individual-level information—could be adapted for use with scRNA-seq datasets. Another approach is to restrict unauthorized access to sensitive datasets, making them available only to researchers who have undergone rigorous vetting and agree to strict data-use agreements. However, this solution is not without its downsides. Restricting access to data can slow the pace of discovery and stifle innovation, particularly for smaller research teams or those in lower-resource institutions, and determining who is ‘trustworthy enough’ to gain access to the data will be challenging. A third, and perhaps more sustainable, solution is to promote greater transparency and education around the risks associated with single-cell data. Just as researchers are trained to handle sensitive clinical data responsibly, they must also be made aware of the privacy risks inherent in scRNA-seq data. By understanding the potential for information leakage, researchers can take proactive steps to minimize the risks—for example, by removing certain types of metadata, applying more rigorous de-identification techniques or limiting the granularity of the data they share. Finally, there is an urgent need for interdisciplinary collaboration between biologists, data scientists, ethicists, legal experts and the public. The privacy challenges posed by scRNA-seq are not unique; they are part of a broader trend in the era of big data, where the boundaries between ‘anonymized and encrypted’ and ‘identifiable’ data are becoming increasingly blurred. By working together, these experts can develop new frameworks and guidelines that balance the need for open data sharing with the imperative to protect individual privacy. Moving forward, it will be important to educate researchers, participants and the public about the risks associated with data sharing and to include clear information about potential risks in informed consent documents. As the power of our analytical tools grows, so too do the risks associated with the data we generate. It is clear that the status quo of research data privacy and confidentiality is no longer sufficient; we must rethink how we collect, store and share biological data to ensure that privacy is protected. However, this should not mean that we should retreat from the principles of open science. On the contrary, the challenges we face should inspire us to innovate. Just as the advent of big data has transformed our ability to understand biology, it can also drive the development of new privacy-preserving technologies and ethical frameworks. By investing in these innovations, we can ensure that the benefits of scRNA-seq are realized without compromising individual privacy. The path forward will require collaboration, creativity and a commitment from researchers, policymakers, funding organizations and participants to building a framework for responsible data sharing and to balancing the needs of science with the rights of individuals, thus enabling scientific discovery while safeguarding the privacy and dignity of individuals. By doing so, the rewards will be immense, not only in terms of scientific discovery, but also in fostering a research environment that is ethical, inclusive and respectful of the individuals whose data make these discoveries possible. In conclusion, while the privacy risks associated with scRNA-seq data are significant, they are not insurmountable. By acknowledging these risks and taking proactive steps to mitigate them, the biomedical research community can continue to push the boundaries of what is possible—safely, responsibly and ethically. None declared. S.S. wrote the first version of the article. All authors significantly contributed to the content and reviewed the manuscript. S.S. received honoraria from AstraZeneca, outside the submitted work, and is a board member of the ERA Immunonephrology Working Group. R.K. is founder and a board member of Sequantrix; a member of the scientific advisory board of Hybridize Therapeutics; received honoraria from Bayer, Chugai, Pfizer, Roche, Genentech, Lilly and GSK; and received research funding from Travere Therapeutics, Galapagos, Novo Nordisk and Ask Bio, outside of the submitted work. V.L. reports royalties as editor of The Kidney, outside the submitted work. S.S., R.K. and V.L. are editorial board members of NDT. V.L. and W.v.B. are board members of the ERA Ethics Committee.

Authors 5

1. Rafael Kramann Aachen Medical Faculty Division of Nephrology and Clinical Immunology Division of Nephrology and Clinical Immunology

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Medical Faculty Division of Nephrology and Clinical Immunology, , Aachen ,

   Division of Nephrology and Clinical Immunology, RWTH Aachen University, Medical Faculty, Aachen, Germany

2. Christoph Kuppe Aachen Medical Faculty Division of Nephrology and Clinical Immunology Division of Nephrology and Clinical Immunology

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Medical Faculty Division of Nephrology and Clinical Immunology, , Aachen ,

   Division of Nephrology and Clinical Immunology, RWTH Aachen University, Medical Faculty, Aachen, Germany

3. Valérie A. Luyckx

   Brigham and Women's Hospital · Harvard University · University of Cape Town · University of Zurich · University Children's Hospital Zurich

   Affiliation as printed

   Brigham and Women's Hospital, Harvard Medical School Renal Division, , Boston, MA ,

   Epidemiology, Biostatistics and Prevention Institute, University of Zurich Department of Public and Global Health, , Zurich ,

   University Children's Hospital, University of Zurich Nephrology Department, , Zurich ,

   University of Cape Town Department of Paediatrics and Child Health, , Cape Town ,

   Department of Paediatrics and Child Health, University of Cape Town, Cape Town, South Africa

   Department of Public and Global Health, Epidemiology, Biostatistics and Prevention Institute, University of Zurich, Zurich, Switzerland

   Nephrology Department, University Children's Hospital, University of Zurich, Zurich, Switzerland

   Renal Division, Brigham and Women's Hospital, Harvard Medical School, Boston, MA, USA

4. Wim Van Biesen

   Ghent University Hospital

   Affiliation as printed

   Ghent University Hospital Renal Division, Department of Internal Medicine, , Ghent ,

   Renal Division, Department of Internal Medicine, Ghent University Hospital, Ghent, Belgium

5. Stefanie Steiger corresponding

   LMU Klinikum · Ludwig-Maximilians-Universität München

   Affiliation as printed

   Hospital of the Ludwig-Maximilians-University Munich Division of Nephrology, Department of Internal Medicine IV, , Munich ,

   Division of Nephrology, Department of Internal Medicine IV, Hospital of the Ludwig-Maximilians-University Munich, Munich, Germany