Extracting replicable associations across multiple studies: Empirical Bayes algorithms for controlling the false...

Area of Science:

• Genomics
• Biomedical Data Analysis
• Statistical Genetics

Background:

• Large-scale biomedical datasets are crucial for identifying genetic associations across numerous fields.
• A key challenge is to reliably extract recurring associations from multiple studies while controlling the false discovery rate.

Purpose of the Study:

• To propose a novel method for the joint analysis of multiple genomic studies.
• To identify associations that replicate across at least k > 1 studies with controlled false discovery rate.

Main Methods:

• Developed several algorithms for joint analysis of multiple studies, modeling study dependencies.
• Proposed SCREEN (Scalable Cluster-based REplicability ENhancement), a three-stage algorithm: study correlation network clustering, within-cluster replicability learning, and cross-cluster result merging.
• Compared proposed algorithms and existing methods using simulated data.

Main Results:

• SCREEN significantly outperformed standard meta-analysis on two real-world datasets.
• Applied to 29 cancer gene expression studies, SCREEN identified consistently up-regulated genes involved in proliferation and cell cycle regulation.
• In a pan-cancer study of HLA complex mutations, SCREEN detected a large module of immune-response genes, identifying thrice more genes than the original study at a similar false discovery rate.

Conclusions:

• SCREEN is a powerful and effective method for identifying replicable genetic associations across multiple studies.
• The algorithm demonstrates high power in detecting biologically relevant gene modules, particularly in complex diseases like cancer.
• SCREEN offers an advancement in controlling false discovery rates for robust cross-study association analysis.