Local false discovery rate estimation with competition-based procedures for variable selection.

Area of Science:

• Statistics
• Bioinformatics
• Genomics

Background:

• Multiple hypothesis testing is crucial for high-dimensional data analysis.
• The false discovery rate (FDR) is a common error control measure.
• Local false discovery rate (fdr) offers individual hypothesis confidence but often requires p-values or known null distributions, which can be unreliable.

Purpose of the Study:

• To propose a new method, TDfdr, for estimating the local false discovery rate (fdr).
• To develop an fdr estimation approach that does not rely on p-values or known null distributions.
• To evaluate TDfdr's performance in high-dimensional data analysis and biomedical applications.

Main Methods:

• Developed TDfdr, a novel approach for fdr estimation using competition-based procedures.
• The method, inspired by knockoff filters, avoids reliance on p-values or known null distributions.
• Validated through extensive simulation studies and applied to real-world biomedical datasets.

Main Results:

• TDfdr accurately estimates the local false discovery rate (fdr) using competition-based procedures.
• The method demonstrated higher discovery power compared to existing popular methods.
• Successfully applied to identify COVID-19 related proteins and HIV-1 drug resistance mutations.

Conclusions:

• TDfdr provides a robust and reliable method for local false discovery rate (fdr) estimation.
• The approach is particularly valuable when p-values or null distributions are unavailable or unreliable.
• TDfdr shows significant promise for advancing high-dimensional data analysis in biomedical research.