Privacy preserving identification of population stratification for collaborative genomic research

Abstract

The rapid improvements in genomic sequencing technology have led to the proliferation of locally collected genomic datasets. Given the sensitivity of genomic data, it is crucial to conduct collaborative studies while preserving the privacy of the individuals. However, before starting any collaborative research effort, the quality of the data needs to be assessed. One of the essential steps of the quality control process is population stratification: identifying the presence of genetic difference in individuals due to subpopulations. One of the common methods used to group genomes of individuals based on ancestry is principal component analysis (PCA). In this article, we propose a privacy-preserving framework which utilizes PCA to assign individuals to populations across multiple collaborators as part of the population stratification step. In our proposed client-server-based scheme, we initially let the server train a global PCA model on a publicly available genomic dataset which contains individuals from multiple populations. The global PCA model is later used to reduce the dimensionality of the local data by each collaborator (client). After adding noise to achieve local differential privacy (LDP), the collaborators send metadata (in the form of their local PCA outputs) about their research datasets to the server, which then aligns the local PCA results to identify the genetic differences among collaborators’ datasets. Our results on real genomic data show that the proposed framework can perform population stratification analysis with high accuracy while preserving the privacy of the research participants.

Genomic privacy preservation in genome-wide association studies: taxonomy, limitations, challenges, and vision

“…For solutions based on differential privacy, it's important to note that even though differential privacy provides high robustness against membership inference attacks [ 44 ]; the most well-known attacks to the GWAS computations [ 45 ]; however, the added noise, necessary to ensure the privacy, comes with a significant reduction in the performances [ 84 ].…”

“…Therefore, Dervishi et al . [ 45 ] focus on performing population stratification analysis by assigning individuals to corresponding populations in a collaborative study manner while maintaining privacy and thususing PCA. The aim is to correctly classify the samples (individuals) that are present in researchers’s local datasets to the corresponding population cluster based on the data sent by the researcher.…”

Creating intelligent cyberinfrastructure for democratizing AI

“…PADI : Within the “privacy, accountability and data integrity” thrust (PADI), early results include novel privacy‐preserving techniques applied to sequential data sharing (Jiang, Yilmaz, and Ayday 2023), the sharing of summary statistics from sensitive databases, and collaborative quality control for research databases (Dervishi et al. 2023). In ongoing research, we are leveraging model cards and model ontologies for accountability AI models and their use.…”

Secure Discovery of Genetic Relatives across Large-Scale and Distributed Genomic Datasets

“…Wang et al (Wang et al 2022) proposed a homomorphic encryption method for identifying genetic relationships across parties, but their approach requires kinship computation for all pairs of samples, which does not scale to large datasets. Other previous approaches(Dervishi et al 2023;Glusman et al 2017;Hormozdiari et al 2014;Robinson and Glusman 2018) rely on sharing a limited amount of processed data between parties to find related samples, which sacrifices both privacy and accuracy to some extent. For instance, Dervishi et al(Dervishi et al 2023) introduced a solution in which the parties reveal a subset of SNPs in a shuffled order for their respective samples to estimate the kinship coefficients.…”

“…Other previous approaches(Dervishi et al 2023;Glusman et al 2017;Hormozdiari et al 2014;Robinson and Glusman 2018) rely on sharing a limited amount of processed data between parties to find related samples, which sacrifices both privacy and accuracy to some extent. For instance, Dervishi et al(Dervishi et al 2023) introduced a solution in which the parties reveal a subset of SNPs in a shuffled order for their respective samples to estimate the kinship coefficients. Robinson andGlusman (2018) andGlusman et al (2017) proposed to compare "fingerprints" obtained by applying a random projection to genomic samples to infer relatedness He et al (2014).…”