حفظ حريم خصوصي كارا در مطالعات وابستگي سطح ژنوم

Significant advancements in genomics have created unprecedented opportunities for personalized medicine an‎d novel therapeutic discoveries. However, this field faces a fundamental conflict: the critical need for large-scale datasets to achieve statistically robust results versus significant privacy challenges that hinder the open sharing of genomic data, one of the most sensitive forms of personal information. This data accessibility bottleneck has become a primary obstacle to scientific progress, particularly for smaller institutions an‎d independent researchers. This thesis addresses this challenge by proposing two novel an‎d complementary solutions to facilitate the responsible use of genomic data. The first solution, the Fed-GWAS framework, is a distributed system for privacy-preserving Genome-Wide Association Studies (GWAS). By integrating cross-device federated learning, blockchain technology for a trust-based incentive mechanism, an‎d a fully collusion-resistant secure aggregation protocol, Fed-GWAS enables individuals to participate in scientific research without sharing their raw data. The framework achieves an exponential reduction in runtime an‎d superior scalability compared to state-of-the-art schemes, enabled by a novel two-layer mining structure. The second solution, the Private Pipeline for Synthetic Data-Sharing (PPSD), tackles the data accessibility problem by enabling the generation of high-fidelity synthetic genomic data. This comprehensive pipeline leverages Generative Adversarial Networks (GANs) an‎d features a custom loss function designed to preserve complex statistical properties an‎d the crucial genotype-phenotype association. A key innovation of PPSD is the pioneering use of Explainable AI (XAI) as a tool for data curators to interpret the model an‎d eva‎luate the privacy risks of the synthetic data before its release. Together, these two frameworks provide a comprehensive approach to the data accessibility crisis in genomics. By empowering both individuals an‎d researchers, this dissertation paves the way for faster, more accurate, an‎d yet responsible an‎d privacy-preserving datadriven discoveries in medical research.

6

مينا عليشاهي

151615