Empirical Bayes ranking and selection methods via semiparametric hierarchical mixture models in microarray studies

Area of Science:

• Genomics
• Biostatistics
• Bioinformatics

Background:

• Microarray studies aim to identify differentially expressed genes for further research.
• Prioritizing genes is crucial due to limited resources in gene screening.
• Existing parametric empirical Bayes methods offer gene ranking based on effect sizes.

Purpose of the Study:

• To develop novel empirical Bayes ranking methods using a semiparametric hierarchical mixture model.
• To improve the selection of candidate genes in microarray studies.
• To explore genes associated with prognosis or disease progression in leukemia.

Main Methods:

• Developed semiparametric hierarchical mixture models for gene ranking.
• Employed a nonparametric prior distribution for effect sizes.
• Utilized the 'smoothing by roughening' approach for prior estimation.

Main Results:

• The proposed semiparametric model allows for effective gene prioritization.
• Information borrowing across genes enhances the separation of differential and non-differential genes.
• The methods were applied to childhood and infant leukemia clinical studies.

Conclusions:

• The semiparametric empirical Bayes approach offers a robust framework for gene selection in microarrays.
• This method improves the identification of significant genes for disease-related research.
• The findings have implications for understanding leukemia prognosis and progression.