Directional false discovery rate control in large-scale multiple comparisons

Area of Science:

• Biostatistics
• Genomics
• Bioinformatics

Background:

• High-throughput technology generates massive gene expression data.
• Identifying differentially expressed genes (under- and over-expressed) is crucial for disease research.
• Existing methods often fail to control false discoveries separately for under- and over-expressed genes.

Purpose of the Study:

• To develop a novel statistical procedure for multiple testing in gene expression analysis.
• To separately control false discovery rates for under-expressed and over-expressed genes.
• To maximize true discoveries while maintaining control over false positives in both directions.

Main Methods:

• A three-classification multiple testing framework was employed.
• A practical, data-driven procedure was developed.
• The procedure was designed to control false discovery rates for under- and over-expressed genes distinctly.

Main Results:

• The proposed procedure is theoretically valid and optimal.
• It maximizes the expected number of true discoveries.
• It simultaneously controls false discovery rates for both under-expressed and over-expressed genes with flexibility in nominal levels.
• Effectiveness demonstrated on two large-scale genomic datasets.

Conclusions:

• The developed procedure offers an effective solution for identifying directional gene expression changes.
• It provides superior control over false discoveries compared to existing methods.
• The flexibility and optimality make it a valuable tool for genomic data analysis.