Simultaneous clustering and estimation of networks in multiple graphical models.

Area of Science:

• Statistics
• Bioinformatics
• Machine Learning

Background:

• Gaussian graphical models are essential for understanding variable dependencies.
• Analyzing multiple populations jointly improves statistical power but existing methods neglect population similarity.
• Clustering populations is often a key objective in multi-population studies.

Purpose of the Study:

• To develop a novel method, SCENT, for simultaneous clustering and network estimation across multiple populations.
• To address limitations of existing methods by accounting for varying population similarities.
• To enable joint learning of population clusters and their graphical structures.

Main Methods:

• Representing precision matrices from multiple populations as a three-way tensor.
• Proposing a low-rank sparse model for joint clustering and network estimation.
• Utilizing a penalized likelihood approach and an augmented Lagrangian algorithm for model fitting.

Main Results:

• SCENT effectively clusters populations and estimates their graphical models simultaneously.
• Theoretical guarantees for clustering accuracy and norm consistency of estimated precision matrices are established.
• Comprehensive simulations demonstrate the method's superior performance.

Conclusions:

• SCENT offers a powerful and flexible framework for multi-population network analysis.
• The method provides valuable insights into population structures and gene coexpression patterns, as shown in the Genotype-Tissue Expression data analysis.
• SCENT advances the joint analysis of multiple graphical models by integrating clustering and estimation.