Privacy-preserving model evaluation for logistic and linear regression using homomorphically encrypted genotype data.

Area of Science:

• Population Genetics
• Cryptography
• Bioinformatics

Background:

• Linear and logistic regression are crucial for analyzing large genetic datasets in population genetics.
• Analyzing sensitive genotype and phenotype data raises significant patient privacy concerns.
• Existing homomorphic encryption methods for secure computation do not fully protect shared model confidentiality.

Purpose of the Study:

• To develop a secure model evaluation method for linear and logistic regression that preserves patient confidentiality.
• To address the limitations of previous cryptographic approaches in protecting shared genetic models.

Main Methods:

• A novel method employing homomorphic encryption for secure model evaluation in population genetics.
• Encryption of input genotypes, output phenotypes, and model parameters for privacy preservation.
• Application to six prediction tasks involving genetic data analysis.

Main Results:

• The proposed method ensures no private information leakage during model inference.
• High accuracy (≥93%) was achieved across all evaluated prediction tasks.
• Inference time was less than ten seconds per individual for approximately 200 genomes.

Conclusions:

• Demonstrates the feasibility of private model evaluation for linear and logistic regression in population genetics.
• Confirms the ability to protect patient confidentiality with theoretical security guarantees.
• Provides an open-source implementation and test data for reproducibility and further research.