A power law global error model for the identification of differentially expressed genes in microarray data

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• High-density oligonucleotide microarrays identify transcriptionally modulated genes.
• Understanding measurement variability is crucial for accurate identification of differentially expressed genes (DEGs).
• The relationship between signal reproducibility and intensity in microarray data requires further clarification.

Purpose of the Study:

• To empirically model the variance-mean dependence in microarray data.
• To improve the performance of existing DEG identification methods.
• To enhance the understanding of intrinsic measurement variability in gene expression studies.

Main Methods:

• Developed a power law global error model (PLGEM) based on empirical data.
• Utilized data from internal and public Affymetrix GeneChip datasets.
• Proposed a novel DEG identification method incorporating model-derived spread estimates and resampling-based hypothesis testing.

Main Results:

• Demonstrated that dispersion of repeated measures follows a power law dependent on the measure's location.
• Constructed a PLGEM applicable to diverse Affymetrix GeneChip datasets.
• Introduced a new DEG identification statistic and algorithm.

Conclusions:

• The new method effectively controls the false positive rate.
• Achieved a favorable trade-off between sensitivity and specificity.
• Provided consistent results regardless of the number of replicates, even with single samples.