Nonparametric estimation of the rediscovery rate.

Area of Science:

• Statistics
• Biostatistics
• Genomics

Background:

• Validation studies are crucial for enhancing the reliability of scientific discoveries and reducing false positives.
• Statistics plays a key role in rigorously quantifying the reproducibility of research findings.
• Two key concepts, rediscovery rate (RDR) and validation false discovery rate (vFDR), have been defined to assess reproducibility.

Purpose of the Study:

• To develop a nonparametric approach for estimating the rediscovery rate (RDR) and validation false discovery rate (vFDR).
• To establish an explicit link between the RDR and the false discovery rate (FDR).
• To explain the difficulties in reproducing statistically significant results, even with a low FDR.

Main Methods:

• Development of a nonparametric statistical method to estimate RDR and vFDR.
• Mathematical derivation of the relationship between RDR and FDR.
• Application of the developed methods to two metabolomics datasets for high-throughput data analysis.

Main Results:

• A novel nonparametric approach for estimating RDR and vFDR is presented.
• An explicit link between RDR and FDR is demonstrated, clarifying reproducibility challenges.
• The methods are successfully illustrated using real-world metabolomics data.

Conclusions:

• The proposed nonparametric approach provides a rigorous way to quantify the reproducibility of scientific findings.
• Understanding the link between RDR and FDR is essential for interpreting the success of validation studies.
• The RDR and vFDR concepts are valuable tools for high-throughput data analysis in fields like metabolomics.