Applying shrinkage variance estimators to the TOST test in high dimensional settings

Area of Science:

• Genomics
• Statistical Bioinformatics
• Computational Biology

Background:

• Identifying differentially expressed genes is common, but identifying constantly expressed genes is crucial yet overlooked.
• Current methods for identifying constantly expressed genes often use naive approaches, leading to high false discovery rates and low statistical power.
• High-dimensional statistical equivalence tests offer a more appropriate approach, but require robust variance estimation.

Purpose of the Study:

• To investigate the impact of shrinkage variance estimators on the two one-sided tests (TOST) for identifying constantly expressed genes in high-dimensional genomic data.
• To address the instability of variance estimators in TOST tests with small sample sizes common in genomics.
• To develop and evaluate an improved statistical framework for gene expression analysis.

Main Methods:

• Simulation studies were conducted to assess the performance of shrinkage variance estimators within the TOST framework.
• Analytic formulas for p-values of the shrinkage variance TOST test were derived.
• The developed method was applied to a real-world genomics dataset.

Main Results:

• Shrinking variance estimators significantly improves the performance of the TOST test in high-dimensional settings.
• The proposed shrinkage variance TOST test demonstrates enhanced statistical power and a reduced false discovery rate.
• Analytic p-value formulas provide a computationally efficient way to apply the method.

Conclusions:

• The application of shrinkage variance estimators to TOST is a valuable advancement for identifying constantly expressed genes.
• This approach offers a more reliable and powerful alternative to traditional methods in high-dimensional genomics.
• The findings have implications for gene function discovery and understanding molecular mechanisms across different conditions.