RDCurve: a nonparametric method to evaluate the stability of ranking procedures

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Reproducibility of gene signatures from high-throughput data, like microarrays, is a significant challenge.
• Existing methods often yield poorly overlapping results and lack biological context in p-values.

Purpose of the Study:

• To introduce a nonparametric method, the ReDiscovery Curve (RDCurve), for estimating the rediscovery frequency of identified gene signatures.
• To provide a robust statistical framework for assessing gene signature reliability and performance.

Main Methods:

• The RDCurve method bootstraps datasets and repeatedly applies ranking procedures to select gene subsets.
• It estimates the probability of rediscovering these selected gene subsets.
• A permutation scheme is used to establish confidence bands under the Null hypothesis for RDCurve significance.

Main Results:

• The RDCurve is nonparametric and model-independent, offering broad applicability.
• It enables assessment of the data's signal-to-noise ratio.
• Allows comparison of different ranking procedures based on expected rediscovery rates and aids in selecting optimal gene set sizes.

Conclusions:

• The RDCurve method offers a reliable approach to evaluate gene signature reproducibility.
• It provides crucial biological context and aids in optimizing gene signature reporting.
• Enhances the interpretability and trustworthiness of findings from high-throughput genomic studies.