A Bayesian calibration model for combining different pre-processing methods in Affymetrix chips.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression studies rely on pre-processing to extract biological signals and manage technical variability.
• The absence of a standardized pre-processing method introduces risks of false positives and negatives in research.
• Researchers must select pre-processing methods, impacting the reliability of differential expression analysis.

Purpose of the Study:

• To develop a robust method for assessing differential gene expression by integrating information from multiple pre-processing techniques.
• To provide a more accurate estimation of 'true' differential expression between biological conditions.
• To demonstrate the utility of a Bayesian approach in gene expression data analysis.

Main Methods:

• Development of a Bayesian calibration model.
• Integration of information from diverse pre-processing methods.
• Estimation of posterior distributions for differential expression values.

Main Results:

• The Bayesian calibration model provides a superior assessment of differential gene expression compared to individual methods.
• The model effectively combines information from multiple pre-processing steps for improved accuracy.
• Demonstrated ability to estimate posterior distributions of differential expression values.

Conclusions:

• The Bayesian calibration model exhibits greater statistical power than individual pre-processing methods on simulated and real data.
• Validation on Affymetrix spike-in data and publicly available experimental data confirms the model's effectiveness.
• The proposed model enhances the biological interpretation and reliability of gene expression studies.