A probe-treatment-reference (PTR) model for the analysis of oligonucleotide expression microarrays

Area of Science:

• Bioinformatics
• Genomics
• Statistical Modeling

Background:

• Microarray data pre-processing involves normalization and summarization to remove variations and estimate transcript abundance.
• Current normalization methods often lack robust strategies for selecting reference arrays.
• Jointly addressing normalization and summarization is crucial for accurate microarray analysis.

Purpose of the Study:

• To develop a novel strategy for reference selection in microarray normalization and summarization.
• To introduce a unified model that integrates normalization and summarization processes.
• To enhance the accuracy and power of detecting differentially expressed genes.

Main Methods:

• Proposed the Probe-Treatment-Reference (PTR) model, allowing multiple references for enhanced normalization and summarization.
• Employed the Least Absolute Deviations (LAD) approach for robust parameter estimation.
• Utilized median polishing for computational implementation of the PTR model.

Main Results:

• The LAD estimator in the PTR model demonstrates robustness with bounded influence.
• The model implicitly identifies an optimal reference for each probe-set.
• Evaluated on Affymetrix spike-in data, the PTR method reduced variations in non-differentially expressed genes.
• Demonstrated increased detection power for differentially expressed genes.

Conclusions:

• The reference effect is a critical factor in microarray pre-processing that requires careful consideration.
• The PTR method provides a flexible framework for addressing reference selection challenges.
• The PTR approach can be integrated with existing normalization algorithms like invariant-set, sub-array, and quantile methods.