Privacy-preserving construction of generalized linear mixed model for biomedical computation

Area of Science:

• Biomedical computation
• Genomic data analysis
• Statistical modeling

Background:

• Generalized linear mixed models (GLMMs) are crucial for analyzing complex biological data, particularly in genome-wide association studies (GWASs).
• Collaborative GWAS across institutions face challenges due to privacy concerns regarding sensitive genomic and health data.
• Existing methods often require data centralization, posing privacy risks.

Purpose of the Study:

• To develop a privacy-preserving Expectation-Maximization (EM) algorithm for collaborative GLMM construction.
• To enable distributed analysis of horizontally partitioned data without data transfer to a central server.
• To facilitate secure collaborative GWAS across multiple institutions.

Main Methods:

• A novel privacy-preserving Expectation-Maximization (EM) algorithm for GLMM construction.
• The algorithm operates on horizontally partitioned data distributed across multiple parties.
• Implementation in R using the rsocket package for secure data communication.

Main Results:

• The collaborative EM algorithm is mathematically equivalent to standard EM algorithms for GLMMs.
• The algorithm demonstrates efficient performance on simulated and real human genomic datasets.
• The developed cGLMM approach is practical for privacy-preserving GLMM construction.

Conclusions:

• The proposed privacy-preserving EM algorithm effectively addresses data privacy concerns in collaborative GLMM construction.
• This method facilitates secure and efficient collaborative analysis of distributed genomic data.
• The open-source R package (cGLMM) is available for practical application.