Power enhancement via multivariate outlier testing with gene expression arrays

Area of Science:

• Genomics
• Bioinformatics
• Statistical genetics

Background:

• Microarray technology is increasingly used in human studies, necessitating robust quality assurance for accurate data interpretation.
• Existing quality control methods may not fully capture complex signal and noise interactions in gene expression data.

Purpose of the Study:

• To develop and validate a formal, dimension-reduction-based approach for microarray quality assessment.
• To improve the inferential power of large-scale clinical studies by identifying and excluding low-quality arrays.

Main Methods:

• A formal approach for microarray quality assessment using dimension reduction of signal and noise measures.
• Parametric multivariate outlier testing applied to expression data.
• Validation using MAQC data, simulated corrupted data, and a large set of human peripheral blood samples.

Main Results:

• The developed method confirmed the absence of outliers in MAQC data at a nominal flagging rate of alpha=0.01.
• A tunable framework was established to assess the sensitivity and specificity of quality assurance criteria.
• Exclusion of arrays identified by the method significantly increased the power to detect differential expression in human peripheral blood samples.

Conclusions:

• The proposed formal approach provides a powerful tool for microarray quality assessment.
• Implementing this quality control strategy enhances the reliability and statistical power of findings in large clinical studies utilizing microarray data.