Inference on the limiting false discovery rate and the p-value threshold parameter assuming weak dependence between...

Area of Science:

• Bioinformatics
• Statistical Genetics
• Genomics

Background:

• Microarray data analysis aims to identify gene expressions associated with disease outcomes.
• Marginal p-values from test statistics are commonly used but require pooling for cross-gene analysis.
• Existing methods may not fully account for dependencies between p-values.

Purpose of the Study:

• To develop a statistical methodology for identifying disease-associated gene expressions from microarray data.
• To model the p-value density as a mixture distribution to improve information pooling across genes.
• To enable estimation and inference on the false discovery rate (FDR) and p-value threshold.

Main Methods:

• Computed test statistics for each gene to generate marginal p-values.
• Modeled the p-value density as a mixture of uniform and scaled normal densities.
• Employed quasi-likelihood estimation under weak dependence assumptions for parameter estimation.
• Applied the methodology to a localized prostate cancer dataset.

Main Results:

• Developed a method for estimating mixture density parameters using quasi-likelihood.
• Enabled asymptotic inference for the false discovery rate (FDR) and p-value threshold.
• Demonstrated the methodology's application on a real-world prostate cancer dataset.
• Assessed performance through simulations.

Conclusions:

• The proposed mixture model and quasi-likelihood approach effectively pool information from microarray data.
• The methodology provides a robust framework for false discovery rate control and gene association studies.
• This approach enhances the identification of disease-related gene expressions in complex biological datasets.