Testing Differential Gene Networks under Nonparanormal Graphical Models with False Discovery Rate Control.

Area of Science:

• Statistics
• Bioinformatics
• Computational Biology

Background:

• Nonparanormal graphical models offer flexibility for non-Gaussian data while retaining interpretability.
• Detecting differential substructure is vital for understanding complex biological networks.

Purpose of the Study:

• To develop a novel statistical method for identifying differential substructure between two nonparanormal graphical models.
• To ensure false discovery rate control in the detection process.

Main Methods:

• A new test statistic is constructed using a truncated estimator for unknown transformation functions.
• Bias-corrected sample covariance is incorporated into the statistic.
• The convergence of the test statistic to its oracle counterpart's distribution is demonstrated.

Main Results:

• The proposed method accurately detects differential substructure in both synthetic and real cancer genomic data.
• The new test statistic exhibits desirable distributional properties.

Conclusions:

• The developed testing framework is simple, scalable, and suitable for large-scale data analysis.
• The R package DNetFinder is available for implementing this computational pipeline.