Learning-Augmented Sketching Offers Improved Performance for Privacy Preserving and Secure GWAS.

Area of Science:

• Computational biology
• Genomics
• Privacy-preserving computation

Background:

• Trusted Execution Environments (TEEs) like Intel SGX enable secure cloud computation.
• Resource limitations in TEEs necessitate memory-efficient methods like sketching for large datasets.
• Existing SkSES method enables privacy-preserving Genome-Wide Association Studies (GWAS) across multiple institutions.

Purpose of the Study:

• To improve the performance and accuracy of the SkSES method for GWAS on large datasets within TEEs.
• To develop a learning-augmented approach that enhances the sketching process for variant identification.
• To achieve higher accuracy in identifying significant Single Nucleotide Polymorphisms (SNPs) while maintaining privacy and memory constraints.

Main Methods:

• Augmenting the SkSES method with a learning-based approach.
• Institutions perform localized, smaller-scale GWAS to identify candidate variant sets.
• Utilizing these candidate sets to guide the sketching process for collective dataset analysis.
• Implementing the method within Trusted Execution Environments (TEEs) for privacy preservation.

Main Results:

• The learning-augmented SkSES method achieves up to a 40% accuracy gain compared to the original SkSES.
• The improved method maintains high accuracy under the same memory constraints.
• Demonstrated effective privacy-preserving GWAS on large, multi-institutional datasets.

Conclusions:

• The learning-augmented SkSES method offers a significant performance improvement for privacy-preserving GWAS in TEEs.
• This approach effectively balances computational efficiency, accuracy, and data privacy.
• The method provides a scalable solution for multi-institutional genetic research without compromising sensitive genotype information.