RPS

RPS computes the Reproducibility Probability Score (RPS) to prioritize differentially expressed genes by estimating the probability that their differential expression signals will reproduce across laboratories using reference microarray datasets from the MAQC project.

Key Features:

• Reproducibility Probability Score: Implements the RPS metric to rank genes by the likelihood their differential expression will reproduce across laboratories.
• Reference interlaboratory variability modeling: Uses reference microarray datasets from the MAQC project to explicitly model interlaboratory measurement variability.
• Mixed-effects models: Fits mixed-effects models to reference data to estimate probe-level correlations and variance components associated with laboratory measurement variability.
• Simulation of hypothetical laboratories: Simulates expression values for hypothetical laboratories consistent with the estimated variability structure given new expression data from a single laboratory.
• User-specified selection rules: Computes, for each gene, the probability of selection under user-specified rules such as P-value thresholds or fold-change criteria.
• Platform-aware reproducibility estimates: Provides quantitative, platform-specific estimates of reproducibility for microarray-based biomarker discovery.

Scientific Applications:

• Prioritization of differentially expressed genes: Ranks genes by their probability of reproducing across independent laboratories to prioritize candidates for validation.
• Validation probability estimation: Estimates the likelihood that a gene selected in one laboratory will be selected in other laboratories using the same microarray platform.
• Comparison of selection methods: Enables assessment of how different selection rules (e.g., P-value thresholds, fold-change criteria) affect reproducibility.
• Reproducible biomarker discovery: Guides selection of biomarkers from microarray experiments by incorporating real estimates of platform-specific variability.
• Empirical reproducibility assessment: Supports empirical comparison showing genes selected by high RPS exhibit improved reproducibility relative to methods that ignore interlaboratory variability.

Methodology:

Fit mixed-effects models to MAQC reference microarray data to estimate probe-level correlations and variance components; given new expression data from a single laboratory, simulate expression values for hypothetical laboratories consistent with the estimated variability structure; compute per-gene probability of selection under a specified selection rule (e.g., P-value threshold, fold-change criterion).