Private detection of relatives in forensic genomics using homomorphic encryption

Area of Science:

• Genomics
• Bioinformatics
• Cryptography

Background:

• Forensic DNA analysis, particularly Single Nucleotide Polymorphisms (SNPs), is crucial for identifying suspects.
• Genomic data is Personal Identifiable Information (PII) protected by strict privacy regulations.
• Homomorphic Encryption (HE) allows computation on encrypted data, preserving privacy.

Purpose of the Study:

• To introduce Homomorphic Encryption (HE)-based methods for privacy-preserving SNP DNA analysis.
• To compute kinship scores for genome queries while maintaining data privacy.
• To evaluate the performance of HE-based methods in a competition setting.

Main Methods:

• Developed three HE-based approaches: one unsupervised and two supervised.
• Applied these methods to compute kinship scores on encrypted SNP data.
• Utilized Intel AVX, Intel HEXL, and Microsoft SEAL libraries for efficient computation.

Main Results:

• Achieved rapid prediction of 400 kinship scores from 2000 encrypted entries within seconds.
• Demonstrated high accuracy (96%-100% auROC score).
• Maintained robust 128-bit security throughout the analysis.

Conclusions:

• HE-based solutions are computationally practical for protecting genomic privacy in Forensic Genetic Genealogy (FGG).
• Enables secure screening of candidate matches for genealogy analysis.
• Facilitates broader data sharing and AI-driven analysis of sensitive genomic information.