Preprocessing of gene expression data by optimally robust estimators.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression data preprocessing involves aggregating raw signal intensities into single expression values.
• Current methods like MAS 5.0 and Illumina's default summarization prioritize high breakdown points over efficiency.
• Robust estimators are widely accepted but their selection is not always driven by efficiency.

Purpose of the Study:

• To describe the application of optimally robust radius-minimax (rmx) estimators for aggregating gene expression data.
• To implement an algorithm for Affymetrix data preprocessing as a variant of MAS 5.0.
• To evaluate the performance of rmx estimators compared to existing methods.

Main Methods:

• Utilized optimally robust radius-minimax (rmx) estimators for signal intensity aggregation.
• Developed an algorithm for Affymetrix data processing, inspired by MAS 5.0.
• Employed datasets from literature and Monte-Carlo simulations, assessing log-normal distribution assumptions via Kolmogorov distance.

Main Results:

• Rmx estimators achieved 10-20% accuracy improvement over Affymetrix MAS 5.0 and 1-5% over Illumina's default method.
• Enhanced reproducibility (Pearson and Spearman correlation) in technical replicates for Affymetrix and most Illumina data.
• Algorithms implemented in the R package RobLoxBioC, available on CRAN.

Conclusions:

• Optimally robust rmx estimators offer high breakdown points and computational feasibility.
• These estimators significantly increase efficiency in bioinformatics procedures.
• Rmx estimators enhance the reproducibility and statistical power of downstream analyses.