Does logarithm transformation of microarray data affect ranking order of differentially expressed genes?

Area of Science:

• Bioinformatics
• Genomics
• Statistical Genetics

Background:

• Microarray analysis commonly employs logarithmic transformation to normalize raw intensity data, aiming for symmetric, Gaussian-like distributions.
• The universal application of this transformation is debated, prompting investigation into its impact on downstream analyses.

Purpose of the Study:

• To assess the effect of logarithmic transformation on identifying differentially expressed genes using various statistical methods.
• To determine if data transformation alters the rank order of significant genes in microarray studies.

Main Methods:

• Comparison of gene significance and rank order using t-test, regularized t-test, and logistic regression on both raw and logarithmically transformed microarray data.
• Analysis of gene concordance and discordance across different statistical tests and significance thresholds.

Main Results:

• Logarithmic transformation led to 20%-40% discordant significant genes, varying by statistical test and threshold.
• The t-test was most sensitive to transformation, followed by the regularized t-test, while logistic regression showed less impact.
• Top-ranked genes (e.g., top 20-50) demonstrated robustness against logarithmic transformation.

Conclusions:

• Data transformation significantly influences the identification of differentially expressed genes, particularly for lower-ranked candidates.
• Researchers must consider the impact of logarithmic transformation on statistical test performance and gene list composition.
• For identifying highly significant genes, the choice of transformation may be less critical.