Inference with approximate local false discovery rates

Area of Science:

• Statistics
• Bioinformatics
• Genetics

Background:

• Efron's 2-group model is standard for large-scale multiple testing, assuming independent test statistics.
• Marginal local false discovery rate (locFDR) controls false discoveries but doesn't account for dependencies.
• Dependent test statistics in realistic settings can increase power, but calculations are often computationally prohibitive.

Purpose of the Study:

• To develop a computationally feasible method to increase power in large-scale multiple testing by accounting for dependent test statistics.
• To introduce and validate the neighborhood local false discovery rate (locFDR_N) for improved statistical decision-making.
• To demonstrate the practical utility of the proposed method in genetic association studies.

Main Methods:

• Proposed using locFDR_N, the probability of a null hypothesis given test statistics in an N-neighborhood.
• Proved optimality of rejecting small locFDR_N within N-neighborhood-guided decisions, showing power increases with N.
• Evaluated computational complexity relative to N, suggesting selection of the largest feasible neighborhood.

Main Results:

• The locFDR_N approach offers substantial power gains over existing practical methods, even with small N-neighborhoods.
• Power increases with the size of the N-neighborhood, balancing computational feasibility.
• Simulations confirmed the proposed method's superior performance.

Conclusions:

• The locFDR_N method provides a powerful and practical approach for large-scale multiple testing with dependent data.
• The method demonstrated significant utility in a real-world genome-wide association study for height.
• This approach offers a valuable tool for researchers dealing with complex, dependent datasets.