An evaluation framework for statistical tests on microarray data.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Microarray analysis is a common functional genomics tool.
• Small sample sizes in experiments lead to unreliable variance estimates.
• New statistical methods are needed for analyzing microarray data from small sample sizes.

Purpose of the Study:

• To evaluate the performance of statistical tests for generating ranked gene lists from two-channel microarray comparisons.
• To assess these methods using data from experiments with limited replicates.

Main Methods:

• Developed an evaluation method using oligonucleotide microarrays with up to 400 replicates per gene.
• Applied Spearman's rank correlation coefficient to compare gene lists from eight statistical tests.
• Tested performance on small random samples derived from the larger dataset.

Main Results:

• Variance stabilizing methods (Cyber-T, SAM, LIMMA) are beneficial for very small sample sizes.
• SAM and the t-test demonstrated superior control of type I error rates.
• With four replicates, all tested methods achieved high correlation with the reference standard.

Conclusions:

• Statistical test selection is crucial for reliable gene ranking in low-replicate microarray studies.
• Variance stabilization and specific tests like SAM and t-test improve data analysis reliability.
• The proposed evaluation method provides a robust benchmark for assessing statistical test performance in functional genomics.