Inferring network structure in non-normal and mixed discrete-continuous genomic data.

Area of Science:

• Genomics
• Statistical Genetics
• Bioinformatics

Background:

• Inferring multivariate interactions among high-dimensional genomic markers is vital for understanding cancer associations.
• Traditional methods like sparse Gaussian graphical models (continuous data) and sparse Ising models (discrete data) have limitations.
• Existing approaches struggle with non-normal continuous data and mixed continuous-discrete data, hindering accurate conditional independence inference.

Purpose of the Study:

• To develop a unified statistical framework for inferring dependence structures in high-dimensional genomic data.
• To address limitations of existing methods when dealing with non-normal continuous data and mixed-type data.
• To enable the inference of sparse conditional independence structures among observed genomic markers.

Main Methods:

• Utilized Gaussian scale mixtures within a unified Bayesian framework.
• Developed methods to handle continuous data with non-normal marginal distributions (e.g., heavy tails, skewness).
• Incorporated a strategy for analyzing mixed continuous and discrete genomic data (e.g., gene expression and mutation status).

Main Results:

• The proposed Gaussian scale mixture framework successfully handles non-normal continuous data.
• The approach effectively analyzes mixed continuous-discrete genomic data, inferring conditional independence among observed variables.
• Simulations demonstrated superior performance compared to alternative techniques.

Conclusions:

• The Gaussian scale mixture approach provides a robust and unified method for dependence structure inference in genomic data.
• This framework overcomes key limitations of traditional graphical models, enabling more accurate analysis of complex cancer genomics data.
• The study highlights the effectiveness of the proposed method through simulations and real-world cancer genomics data analysis.