Integrating hypertension phenotype and genotype with hybrid non-negative matrix factorization

Area of Science:

• Computational biology
• Genetics
• Machine learning

Background:

• Hypertension is a complex condition requiring better patient subtyping for targeted treatments.
• Current machine learning methods struggle to integrate phenotype and genotype data effectively for clinical interpretability.

Purpose of the Study:

• To develop an interpretable machine learning model for patient stratification using both phenotypic and genetic data.
• To identify hypertension subtypes with shared pathophysiologic mechanisms for more uniform treatment responses.

Main Methods:

• A novel hybrid non-negative matrix factorization (HNMF) method was developed to integrate phenotype and genotype data.
• HNMF uses distinct loss functions (Frobenius for phenotype, KL for genotype) for simultaneous matrix approximation.
• An alternating projected gradient method was employed to solve the approximation problem.

Main Results:

• Simulations demonstrated HNMF's fast convergence and accuracy in estimating factor matrices.
• On a clinical dataset, HNMF successfully stratified patients and identified associations with cardiac mechanics.
• HNMF significantly outperformed six other models and three recent integrative clustering methods.

Conclusions:

• HNMF provides an effective approach for integrating diverse patient data for improved stratification.
• The method reveals intuitive phenotype-genotype interactions, offering insights into cardiac abnormalities.
• HNMF offers a valuable tool for precision medicine in hypertension management.