Statistical methods for chip calibration and saturation effects in antibody-spiked gene expression data

Area of Science:

• Genomics
• Bioinformatics
• Statistical Modeling

Background:

• Oligonucleotide microarrays enable high-throughput gene expression analysis.
• Antibody spiking is used to establish expression curves for detecting expression limits.
• Expression curves vary across chips and are prone to saturation, necessitating standardization.

Purpose of the Study:

• To introduce non-parametric methods for standardizing gene expression curves using univariate smoothers.
• To explore parametric methods for efficient analysis of standardized curves.
• To present an improved parametric analysis using a weighted linear mixed effects model that avoids data exclusion.

Main Methods:

• Non-parametric standardization using univariate smoothers.
• Parametric analysis of standardized curves.
• Weighted linear mixed effects modeling for curve analysis without data truncation.

Main Results:

• Developed and evaluated non-parametric and parametric methods for expression curve standardization.
• The weighted linear mixed effects model demonstrated significantly more accurate predictions than naive linear extrapolation.
• Simulations confirmed the effectiveness of both proposed methodologies.

Conclusions:

• Non-parametric and parametric approaches offer robust standardization for gene expression curves.
• Weighted linear mixed effects models provide a superior alternative to traditional methods for analyzing saturated expression data.
• These methods enhance the accuracy and reliability of gene expression analysis from oligonucleotide microarrays.