MixTwice: large-scale hypothesis testing for peptide arrays by variance mixing.

Area of Science:

• Immunoproteomics
• Statistical Bioinformatics

Background:

• Peptide microarrays are crucial for measuring antibody abundance in patient serum.
• High-dimensional data and small sample sizes challenge traditional statistical methods, including false discovery rate (FDR) control.

Purpose of the Study:

• To introduce MixTwice, an empirical Bayesian tool addressing limitations in current statistical approaches for peptide microarray analysis.
• To enhance the reproducibility and statistical power of identifying significant peptide markers.

Main Methods:

• MixTwice estimates local FDR and local false sign rates using data on estimated effects and standard errors.
• It employs two mixing distributions for underlying effects and variance parameters.
• Model fitting utilizes constrained optimization with weak shape constraints, such as unimodality.

Main Results:

• Numerical experiments demonstrate MixTwice's accuracy in estimating generative parameters.
• The tool effectively identifies non-null peptides, showing improved power.
• In a rheumatoid arthritis study, MixTwice identified relevant peptide markers with weak signals and exhibited strong reproducibility.

Conclusions:

• MixTwice offers a powerful and reproducible statistical framework for peptide microarray data analysis in immunoproteomics.
• The tool overcomes challenges posed by high dimensionality and small sample sizes.
• MixTwice advances the field by providing robust methods for FDR and false sign rate control.