MixTwice

MixTwice applies empirical Bayes methods to peptide microarray data to compute local false discovery rate and local false sign rate statistics for improved hypothesis testing in high-dimensional, small-sample experiments.

Key Features:

• Empirical Bayesian framework: Computes local FDR and local false sign rate statistics using empirical Bayes techniques for large-scale hypothesis testing.
• Dual mixing distribution estimation: Estimates two mixing distributions separately: one for underlying effects and one for underlying variance parameters.
• Constrained optimization: Fits mixing distributions under weak shape constraints using constrained optimization, including enforcing unimodality of the effect distribution.
• Generative parameter estimation: Provides accurate estimation of generative parameters for the assumed mixture model.
• Power for weak signals: Enhances identification of non-null peptides even when signals are weak.
• Implementation: Provided as an R package for integration into computational workflows.
• Target data characteristics: Specifically addresses high dimensionality and small sample sizes typical of peptide microarray experiments.

Scientific Applications:

• Peptide microarray hypothesis testing: Controls false discovery measures in large-scale peptide microarray experiments.
• Antibody marker discovery: Identifies antibody abundance markers in patient serum samples from peptide arrays.
• Immunoproteomics studies: Applied to immunoproteomics analyses, including peptide array studies of rheumatoid arthritis.
• FDR control in proteomics: Improves reproducibility and statistical power for FDR-controlled discovery in high-dimensional proteomic data.

Methodology:

Empirical Bayesian estimation of local FDR and local false sign rate; estimation of two mixing distributions (effects and variance parameters); constrained optimization to fit mixing distributions under weak shape constraints such as unimodality of the effect distribution.