Double Empirical Bayes Testing.

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Analyzing large-scale, multi-experiment studies presents challenges in assessing individual experiments and overall results.
• Existing methods may struggle with high-dimensional covariate data common in modern scientific studies.

Purpose of the Study:

• To develop a novel empirical Bayes method, Double Empirical Bayes Testing (DEBT), for analyzing multi-experiment studies with numerous covariates.
• To enhance statistical power and control the false discovery rate (FDR) in complex experimental data.

Main Methods:

• DEBT employs a two-stage empirical Bayes approach, building on Efron's (2008) testing framework.
• Stage 1 utilizes a deep neural network prior to identify significant experiments.
• Stage 2 employs an empirical Bayes knockoff filter to select covariates predictive of experimental significance.

Main Results:

• DEBT demonstrated increased discovery of significant outcomes in both simulated and real data.
• The method effectively selects more features, particularly when underlying signals are weak.
• In a cancer cell line study, DEBT identified biologically plausible genomic drivers of drug response.

Conclusions:

• DEBT offers a powerful and robust framework for multi-experiment analysis with high-dimensional covariates.
• The method enhances the identification of significant findings and relevant biological drivers in complex datasets.