Private Epigenetic PaceMaker Detector Using Homomorphic Encryption - Extended Abstract

Privacy Preserving Epigenetic PaceMaker Stronger Privacy and Improved Efficiency

“…We presented a new privacy preserving protocol for computing the EPM model over federated data. Our protocol offers better asymptotic communication complexity and stronger privacy than the prior work [16] as well as faster concrete efficiency (see Table 1). The uniqueness of the direction pursued here stems from the fact that privacy in the context of epigenetics, as opposed to genome wide association, was rarely studied before, only a single work handling gene expression data [3] preceded our direction of epigenetic aging as pursued here.…”

“…Being offline throughout the bulk of the computation makes it considerably easier to safeguard this server, thus increasing the plausibility of the non-collusion assumption. 2 We implemented our protocol and show that it offers a substantial performance speed-up over the implementation of [16]: securely computing the EPM from 716 methylation sites in 3 hours (cf. 24 sites in 3 hours in [16]).…”

“…2 We implemented our protocol and show that it offers a substantial performance speed-up over the implementation of [16]: securely computing the EPM from 716 methylation sites in 3 hours (cf. 24 sites in 3 hours in [16]). These results can serve as a pilot for data security measures integrated in vast medical applications where personal privacy is imperative.…”

“…Prior work on privacy preserving genome analysis using homomorphic encryption focused primarily on Genome Wide Association (GWAS) [24,32,8,7,12] -i.e., statistically associating innate genome variability in single nucleotide polymorphism (SNPs) with a risk for a disease or a particular trait-as well as on privacy preserving classification of DNA and RNA sequences of tumor tissues [20,10] and viral strains [41,2] respectively. Recently, privacy preserving epigenetics -i.e., the study of how behavior and environmental factors lead to genome changes that in turn affect the phenotype-was considered in [3,16], analysing gene expression data [3] and DNA methylation data [16].Elaborating on the latter, DNA methylation are chemical changes in the genome that are linked to numerous developmental, physiologic, and pathologic processes including malignancy, infections, and aging. A breakthrough in this area was achieved by the epigenetic clock of Horvath [21] that predicted human chronological age based on observing several hundreds of methylation sites along the genome.…”

“…[16], proposing a privacy preserving protocol for computing the EPM model using FHE as a central tool. Their protocol required two servers throughout the entire computation, resulting in two main drawbacks: first, their protocol is insecure if an adversary corrupts both servers; second, their protocol suffers from a high communication and computational toll, making it impractical to real data volumes.In this work we propose a new privacy-preserving protocol for the EPM, improving over the prior work [16] in the following aspects:-Improved complexity: no communication. In [16], secure computing entails multiple rounds of communication between the two servers: one communication round for each expectation-maximization iteration in computing the EPM, transmitting O(nm) ciphertexts in each iteration (for n the number of methylation sites, and m the number of individuals in the dataset).…”