Using of normalizations for gene expression analysis

Area of Science:

• Bioinformatics
• Statistical genetics
• Computational biology

Background:

• Gene expression data normalization is crucial for removing systematic variations.
• However, using normalized data for hypothesis testing raises concerns about statistical validity.
• Existing normalization techniques may affect the reliability of multiple testing procedures.

Purpose of the Study:

• To evaluate the impact of common gene expression data normalization methods on hypothesis testing.
• To determine if quantile, global, and delta-sequence normalizations preserve the nominal level of the Bonferroni multiple testing procedure.
• To assess the risk of false positives when using normalized data in statistical inference.

Main Methods:

• Simulating gene expression data with correlation structures mirroring real biological data.
• Applying quantile, global, and delta-sequence normalization techniques to the simulated datasets.
• Assessing the performance of the Bonferroni multiple testing procedure on both normalized and unnormalized data.

Main Results:

• The study found that all tested normalization methods (quantile, global, delta-sequence) can lead to a loss of the nominal significance level.
• This loss increases the rate of false positives, potentially leading to incorrect conclusions in hypothesis testing.
• The extent of this issue varies depending on the specific normalization method and data characteristics.

Conclusions:

• Caution is advised when using normalized gene expression data for hypothesis testing, particularly with Bonferroni correction.
• The choice of normalization method can significantly influence the Type I error rate.
• Further research is needed to develop robust normalization strategies that maintain statistical integrity for downstream analyses.