Statistical Assessment of Depth Normalization for Small RNA Sequencing.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Depth normalization is crucial for accurate analysis of sequencing data, particularly small RNA sequencing.
• Existing assessment methods rely on simulated or limited cell-line data, lacking real-world applicability.
• A need exists for robust benchmark datasets to objectively evaluate normalization strategies.

Purpose of the Study:

• To develop and provide benchmark data and computational tools for assessing depth normalization methods in small RNA sequencing.
• To evaluate the performance of existing normalization methods using a novel, realistic benchmark dataset.
• To understand how differential expression patterns influence normalization method efficacy.

Main Methods:

• Collected paired microRNA sequencing data from tumor samples under uniform and non-uniform handling.
• Developed a data perturbation algorithm to generate additional simulated dataset pairs.
• Assembled computational tools for visualizing and quantifying the performance of normalization methods.

Main Results:

• Validated the quality of the benchmark data and confirmed the necessity of normalization for the test data.
• Assessed nine normalization methods, identifying trimmed mean of M-values as a superior scaling method.
• Found that median and upper quartile methods performed poorly, while 'remove unwanted variation' showed trade-offs between sensitivity and specificity.

Conclusions:

• Provides essential benchmark data and tools for depth normalization assessment in small RNA sequencing.
• Demonstrates that normalization method performance is contingent on the specific patterns of differential gene expression.
• Emphasizes the ongoing need for research to develop more effective normalization strategies for sequencing data analysis.