Searching across-cohort relatives in 54,092 GWAS samples via encrypted genotype regression

Qixin Zhang, Tianzi Liu, Xinxin Guo, Jianxin Zhen, Mengyuan Yang, Saber Khederzadeh, Fang Zhou, Xiaotong Han, Qiwen Zheng, Peilin Jia, Xiaohu Ding, Mingguang He, Xin Zou, Jia-Kai Liao, Hongxin Zhang, Ji He, Xiaofeng Zhu, Daru Lu, Hongyan Chen, Changqing Zeng, Fan Liu, Hou‐Feng Zheng, Siyang Liu, Haiming Xu, Guo‐Bo ChenbioRxiv 2022 · DOI

TL;DR

This paper introduces encG-reg, a novel method for identifying genetic relatives across different research cohorts using encrypted genotype data. This approach overcomes privacy concerns associated with sharing individual-level genetic information, enabling accurate relatedness estimation without compromising data confidentiality.

What Problem Does This Paper Address?

Genomic studies often involve multiple, distributed cohorts. Identifying genetic relatives across these cohorts is crucial for various analyses, such as controlling for population stratification in Genome-Wide Association Studies (GWAS) or understanding familial structures. However, sharing individual-level genotype data for such analyses is hampered by privacy and ethical constraints. Existing methods that rely on summary statistics or homomorphic encryption can be computationally intensive or lack precision. This paper addresses the challenge of performing accurate, privacy-preserving relatedness estimation across cohorts.

Methods

The core innovation is encG-reg, a regression-based approach that uses encrypted genotype data. The encryption is achieved by multiplying the original genotype matrix of each cohort by a randomly generated matrix (akin to a private key). This process, rooted in random matrix theory, masks the original data while preserving its precision. The method derives the expectation and sampling variance for encG-reg, which are shown to be comparable to traditional methods that require raw data. The authors also establish theoretical bounds for the number of markers (me) and the dimension of the random matrix (k) needed for accurate relatedness detection, balancing precision with computational cost.

Datasets

The authors validated encG-reg on two large-scale datasets:

  • UK Biobank (UKB): A multi-ethnicity dataset comprising 485,158 individuals from 19 assessment centers. This dataset was used to test both exhaustive and parsimony designs for SNP selection.
  • Chinese Cohorts: A collaboration involving 9 Chinese datasets from 5 research institutes, totaling 54,092 individuals. This multi-center application served as a real-world test of encG-reg in complex, diverse environments.

Key Insights and Ideas

  • Privacy-Preserving Relatedness is Achievable: The study demonstrates that it's possible to accurately identify genetic relatives across cohorts without ever exposing raw individual genotype data, a significant step for collaborative genomics.
  • Random Matrix Theory as a Cryptographic Tool: The paper elegantly leverages random matrix theory to mask genotype data, showing its practical application beyond theoretical statistics.
  • Balancing Precision and Privacy: encG-reg offers a tunable precision by adjusting parameters like the number of markers (me) and the random matrix dimension (k), allowing researchers to tailor the method to their specific needs and privacy requirements.
  • Inter-cohort Relatedness is Feasible: The successful application across diverse Chinese cohorts, with varying genotyping platforms and ethnic backgrounds, highlights the robustness of encG-reg for real-world, large-scale genomic collaborations.

Key Results

  • encG-reg achieved accuracy comparable to KING, a widely used software that requires raw genotype data, for detecting relatedness up to the 2nd degree in the UK Biobank.
  • In the Chinese multi-center study with 54,092 samples, encG-reg successfully identified true relatives across cohorts with different ethnic backgrounds and genotypic qualities.
  • The method demonstrated well-controlled Type I and Type II error rates, crucial for reliable genetic analyses.
  • The computational time complexity was found to be manageable for biobank-scale data.

Limitations and Future Work

  • Additive GRM Focus: The current study focused on additive genetic relationship matrices (GRM). The authors suggest extending encG-reg to handle dominance-GRM for more nuanced relatedness detection (e.g., distinguishing parent-offspring from full sibs more precisely).
  • Distant Relatives: While the study demonstrated detection of 1st and 2nd-degree relatives, further examination for inferring more distant relatives would require datasets with more detailed pedigree information and more comprehensive comparisons.
  • Potential Attacks: While the authors are confident in the security of encG-reg, they acknowledge potential adversary attacks (e.g., PCA-attack) on the central site. They suggest using SNPs in linkage equilibrium as a defense mechanism.
  • Homomorphic Encryption Comparison: The paper briefly mentions homomorphic encryption (HE) as an alternative but notes its substantial computational cost. Further comparative studies on the trade-offs between encG-reg and HE for specific applications could be valuable.

Summary updated

Abstract

Abstract Explicitly sharing individual level data in genomics studies has many merits comparing to sharing summary statistics, including more strict QCs, common statistical analyses, relative identification and improved statistical power in GWAS, but it is hampered by privacy or ethical constraints. In this study, we developed encG-reg , a regression approach that can detect relatives of various degrees based on encrypted genomic data, which is immune of ethical constraints. The encryption properties of encG-reg are based on the random matrix theory by masking the original genotypic matrix without sacrificing precision of individual-level genotype data. We established a connection between the dimension of a random matrix, which masked genotype matrices, and the required precision of a study for encrypted genotype data. encG-reg has false positive and false negative rates equivalent to sharing original individual level data, and is computationally efficient when searching relatives. We split the UK Biobank into their respective centers, and then encrypted the genotype data. We observed that the relatives estimated using encG-reg was equivalently accurate with the estimation by KING, which is a widely used software but requires original genotype data. In a more complex application, we launched a finely devised multi-center collaboration across 5 research institutes in China, covering 9 cohorts of 54,092 GWAS samples. encG-reg again identified true relatives existing across the cohorts with even different ethnic backgrounds and genotypic qualities. Our study clearly demonstrates that encrypted genomic data can be used for data sharing without loss of information or data sharing barrier. Author Summary Estimating pairwise genetic relatedness within a single cohort is straightforward. However, in practice, related samples are often distributed across different cohorts, making it challenging to estimate inter-cohort relatedness. In this study, we propose a method called encrypted genotype regression ( encG-reg ), which provides an unbiased estimation of inter-cohort relatedness using encrypted genotypes. The genotype matrix of each cohort is masked by a random matrix, which acts similarly to a private key in a cryptographic scheme. This masking process produces encrypted genotypes, which are a projection of the original genotype matrix. We derive the expectation and particularly the sampling variance for encG-reg , the latter involves eighth-order moments calculation. encG-reg allows us to accurately identify relatedness across cohorts, even for large-scale biobank data. To demonstrate the efficacy of encG-reg , we verified it in a multi-ethnicity UK Biobank dataset comprising 485,158 samples. For this case, we successfully tracked down to the 1st-degree relatedness (such as full sibs and parent-offspring). Furthermore, we used encG-reg in a collaboration involving 9 Chinese cohorts, encompassing a total of 54,092 samples from 5 genomic centers. It is worth noting that if the number of effective markers is sufficient encG-reg has the potential to detect even more distant degrees of relatedness beyond what we demonstrated.
BibTeX
@misc{zhang2022searching,
  author = {Qixin Zhang and Tianzi Liu and Xinxin Guo and Jianxin Zhen and Mengyuan Yang and Saber Khederzadeh and Fang Zhou and Xiaotong Han and Qiwen Zheng and Peilin Jia and Xiaohu Ding and Mingguang He and Xin Zou and Jia-Kai Liao and Hongxin Zhang and Ji He and Xiaofeng Zhu and Daru Lu and Hongyan Chen and Changqing Zeng and Fan Liu and Hou‐Feng Zheng and Siyang Liu and Haiming Xu and Guo‐Bo Chen},
  title = {{Searching across-cohort relatives in 54,092 GWAS samples via encrypted genotype regression}},
  year = {2022},
  doi = {10.1101/2022.10.18.512407},
  howpublished = {bioRxiv},
}
Figure