Replicability analysis in genome-wide association studies via Cartesian hidden Markov models

Area of Science:

• Genetics
• Statistical genomics
• Bioinformatics

Background:

• Replicability analysis is crucial for validating scientific findings, especially in genome-wide association studies (GWAS).
• High correlation among neighboring single nucleotide polymorphisms (SNPs) necessitates methods that exploit this local dependence structure.
• Existing methods may not fully leverage the intricate relationships between adjacent SNPs.

Purpose of the Study:

• To develop a novel multiple testing procedure for replicability analysis across two studies.
• To introduce a method that effectively utilizes the dependency information among adjacent SNPs.
• To enhance the accuracy and efficiency of detecting replicated genetic signals.

Main Methods:

• Proposed a novel multiple testing procedure named repLIS.
• Utilized a Cartesian hidden Markov model (CHMM) to characterize local dependence structure among adjacent SNPs.
• Employed a four-state Markov chain to model SNP dependencies.

Main Results:

• The repLIS procedure controls the false discovery rate (FDR) at the nominal level α.
• Demonstrated theoretical optimality with the smallest false non-discovery rate (FNR).
• Simulation studies and real data analyses showed repLIS is valid and more efficient than Benjamini-Hochberg (BH) and repfdr methods.

Conclusions:

• The repLIS procedure effectively controls FDR in replicability analysis.
• Leveraging SNP dependency information enhances the efficiency of detecting replicated signals.
• repLIS offers a statistically sound and more efficient approach for genetic replicability studies.