Joint Estimation of Precision Matrices in Heterogeneous Populations

Area of Science:

• Statistics
• Computational Biology
• Genomics

Background:

• Estimating precision matrices is crucial for understanding complex data structures.
• Existing methods struggle with heterogeneous populations and high-dimensional data.
• Identifying subpopulation structures is key for accurate matrix estimation.

Purpose of the Study:

• To develop a general framework for estimating precision matrices in heterogeneous populations.
• To improve the accuracy and applicability of precision matrix estimation methods.
• To handle complex, high-dimensional datasets, including biological data.

Main Methods:

• A Laplacian shrinkage penalty is employed to encourage similarity among estimates from related subpopulations.
• An efficient alternating direction method of multipliers (ADMM) algorithm is proposed for parameter estimation.
• A hierarchical clustering-based Laplacian penalty is introduced for unknown population structures.

Main Results:

• The proposed method establishes variable selection and norm consistency for various distributions.
• An extension for high-dimensional data identifies joint block diagonal structures.
• Numerical studies and cancer gene expression data analysis demonstrate advantages over existing methods.

Conclusions:

• The framework provides a robust approach for precision matrix estimation in heterogeneous settings.
• The method is applicable to complex biological data, offering insights into distinct cancer subtypes.
• The data-driven hierarchical clustering approach enhances applicability for unknown population structures.