A practical guide to methods controlling false discoveries in computational biology.

Area of Science:

• Statistical methodology
• Computational biology
• Bioinformatics

Background:

• High-throughput studies test millions of hypotheses, necessitating robust error control.
• False discovery rate (FDR) control is crucial for managing statistical errors in large-scale testing.
• Classic FDR methods rely solely on p-values, while modern methods incorporate covariates for enhanced power.

Purpose of the Study:

• To systematically benchmark and compare the accuracy, applicability, and ease of use of classic and modern FDR-controlling methods.
• To evaluate the performance of FDR methods with and without informative covariates.
• To provide practical guidance for researchers selecting FDR methods.

Main Methods:

• Simulation studies were conducted to assess method performance under various conditions.
• Six case studies in computational biology were used to evaluate real-world applicability.
• Two classic and six modern FDR-controlling methods were benchmarked.

Main Results:

• Modern FDR methods incorporating covariates showed modest power improvements over classic methods.
• Most methods effectively controlled the FDR, with two modern methods showing issues in specific settings.
• Performance gains increased with covariate informativeness, total tests, and proportion of non-null hypotheses.

Conclusions:

• Modern FDR methods utilizing informative covariates offer advantages for hypothesis testing.
• The relative benefits of modern methods depend on the specific application and covariate quality.
• Recommendations are provided to assist researchers in choosing appropriate FDR correction strategies.