Genotype Extraction and False Relative Attacks: Security Risks to Third-Party Genetic Genealogy Services Beyond Identity Inference

Abstract

Customers of direct-to-consumer (DTC) genetic testing services routinely download their raw genetic data and give it to third-party companies that support additional features. One type of analysis, called genetic genealogy, uses genetic data and genealogical methods to find new relatives. While genetic genealogy is quite popular, it has raised new privacy concerns. Genetic genealogy services can be leveraged to find the person corresponding to anonymous genetic data and have been used dozens of times by law enforcement to solve crimes. We hypothesized that the open design and broad API offered by some genetic genealogy services raise other significant security and privacy issues. To test this hypothesis, we analyzed the security practices of GEDmatch, the largest third-party genetic genealogy service. Here, we experimentally show how the GEDmatch API is vulnerable to a number of attacks from an adversary that only uploads normally formatted genetic data files and runs standard queries. Using a small number of specifically designed files and queries, an attacker can extract a large percentage of the genetic markers from other users; 92% of markers can be extracted with 98% accuracy, including hundreds of medically sensitive markers. We also find that an adversary can construct genetic data files that falsely appear like relatives to other samples in the database; in certain situations, these false relatives can be used to make the re-identification of genetic data more difficult. These attacks are possible because of the rich set of features supported by the API, including detailed visualizations, that are meant to enhance usability. We conclude with security recommendations for genetic genealogy services.

Differential Privacy Protection Against Membership Inference Attack on Machine Learning for Genomic Data

“…The leakage of genetic privacy has been frequently reported where the identity or sensitive trait of a particular individual was disclosed by the identity tracing attacks, 30,31 attribute disclosure attacks via DNA, 4 and completion attacks. 32,33 Several privacy-preserving solutions for genomic data analytics have been proposed, including DP strategies for genome-wide association studies (GWAS), 7,34 crytographic solutions including homomorphic encryption 12,35 and secure multi-party computation (MPC) solutions. 35 However, these privacy-preserving methods aim to mitigate privacy leakage when data is shared and none of them have considered privacy leakage when sharing models.…”

Genetic information insecurity as state of the art

“…Data sharing is dependent upon the data controller's wants and needs, barring any legal or business requirements from other involved stakeholders. Genetic database vulnerabilities have been well-studied and disclosed (Edge and Coop, 2020;Ney et al, 2020;Naveed et al, 2015;Erlich and Narayanan, 2014;Gymrek et al, 2013). For example, the contents of the entire GEDmatch database could be leaked by uploading artificial genomes (Ney et al, 2020).…”

“…Genetic database vulnerabilities have been well-studied and disclosed (Edge and Coop, 2020;Ney et al, 2020;Naveed et al, 2015;Erlich and Narayanan, 2014;Gymrek et al, 2013). For example, the contents of the entire GEDmatch database could be leaked by uploading artificial genomes (Ney et al, 2020). Such an attack would violate the confidentiality of more than a million users' and their relatives' genetic data because the information is not truly anonymized.…”

Genetic Information Insecurity as State of the Art

“…Genetic database vulnerabilities have been well studied and disclosed ( Gymrek et al, 2013 ; Erlich and Narayanan, 2014 ; Naveed et al, 2015 ; Edge et al, 2017 ; Ney et al, 2018 ; Vinatzer et al, 2019 ; Edge and Coop, 2020 ; Ney et al, 2020 ). For example, the contents of the entire GEDmatch database could be leaked by uploading artificial genomes ( Ney et al, 2020 ). Such an attack would violate the confidentiality of more than a million users’ and their relatives’ genetic data because the information is not truly anonymized.…”