Coconut: covariate-assisted composite null hypothesis testing with applications to replicability analysis of...

Area of Science:

• Biostatistics
• Genomics
• Computational Biology

Background:

• Multiple testing of composite null hypotheses is crucial for detecting simultaneous signals across studies.
• Incorporating external information into simple null hypotheses is common, but leveraging auxiliary covariates for composite null hypotheses to enhance statistical power remains difficult.

Purpose of the Study:

• To introduce a robust and powerful covariate-assisted composite null hypothesis testing (CoCoNuT) procedure.
• To identify replicable signals across studies by controlling the false discovery rate (FDR) using a Bayesian framework.
• To exploit auxiliary covariates to improve statistical power in hypothesis testing.

Main Methods:

• Developed the covariate-assisted composite null hypothesis testing (CoCoNuT) procedure using a Bayesian framework.
• Employed a three-dimensional mixture model integrating two primary studies and an auxiliary covariate.
• Utilized the local false discovery rate to capture cross-study and cross-feature information, accounting for study heterogeneity.

Main Results:

• CoCoNuT effectively identifies replicable signals across studies while asymptotically controlling the false discovery rate.
• The method optimally captures cross-study and cross-feature information, leading to improved feature importance rankings.
• CoCoNuT demonstrates superior performance compared to methods that do not utilize auxiliary covariates.

Conclusions:

• CoCoNuT is a valid and efficient procedure for composite null hypothesis testing.
• The method shows higher statistical power in identifying replicable genetic variants, as demonstrated in schizophrenia genome-wide association studies.
• CoCoNuT successfully leverages auxiliary studies to enhance the discovery of significant findings.