Using Oblivious RAM in Genomic Studies

Abstract

Since the development of tree-based Oblivious RAMs by Shi et al. it has become apparent that privacy preserving outsourced storage can be practical. Although most current constructions follow a client-server model, in many applications, such as Genome Wide Association Studies (GWAS), it is desirable that multiple entities can share data, while being able to hide access patterns not only from the server, but also from any other entities that can access parts of the data. Inspired by the efficiency and simplicity of Path-ORAM, in this work, we study an extension of Path-ORAM that allows oblivious sharing of data in a multi-client setting, so that accesses can be hidden from the server and from other clients. We address various challenges that emerge when using Path-ORAM in a multi-client setting, and prove that with adequate changes, Path-ORAM is still secure in a setting, where the clients are semi-honest, do not trust each other, but try to learn the access patterns of each other. We demonstrate our ORAM construction in a GWAS setting. Our experiments show that in databases storing $2^{23}$ data blocks (corresponding to a database holding $2^{17}$ blocks per client, capable of storing human genome in the form of SNPs, for 100 clients), the average query time is less than 7 s, yielding a secure and practical solution.