A gene selection method for GeneChip array data with small sample sizes.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Microarray experiments with small sample sizes present challenges in accurate p-value estimation and gene selection.
• Permutation-based methods, while sensitive, yield discrete p-values in small samples, hindering false discovery rate (FDR) control.
• Existing methods struggle with non-uniform p-value distributions from true null hypotheses in small datasets.

Purpose of the Study:

• To develop a novel method for robust p-value estimation and gene selection in small sample size microarray experiments.
• To address the limitations of permutation-based methods and improve the reliability of statistical inference.
• To provide a more powerful and accurate approach for identifying significant genes.

Main Methods:

• Proposed a model-based information sharing (MBIS) method utilizing gene-wise information sharing after data transformation.
• Employed a normal distribution model to estimate mean differences for true null hypotheses across experimental conditions.
• Calculated uniformly distributed p-values from true nulls based on the estimated model parameters.

Main Results:

• The MBIS method generates uniformly distributed p-values from true nulls, facilitating more reliable FDR control.
• The method effectively utilizes information shared among genes to enhance statistical power.
• Gene selection based on cutoff p-values is performed, followed by false discovery rate estimation.

Conclusions:

• Simulation studies and real data analysis demonstrate MBIS superiority over existing methods in power and reliability.
• The MBIS method offers a robust solution for gene selection in small sample size microarray studies.
• The approach has broad applicability across various experimental scenarios in genomics research.