Privacy-Preserving Viral Strain Classification Through a Client-Server Application Using An Open-Source Fully Homomorphic Encryption Library

Abstract

ML techniques and outsourcing are being increasingly used by researchers in their efforts to look into and better understand SARS-CoV-2 and combat the spread of the virus. However, this brings about privacy issues that surround the sharing of, and training of ML models on, SARS-CoV-2 genomic sequences and contextual data, potentially leading to the reidentification of the owners of such genomic data. Thus, there is a need to develop methods of protecting patients’ privacy, all while allowing researchers and medical professionals to continue the use of ML techniques and outsourcing to make better informed medical decisions and take more effective actions against the spread of the virus. To that end, this paper proposes a fully homomorphic encryption-based viral classification framework and logistic regression model based on Concrete-ML, a fully open-source FHE ML library.