DETECTING MULTIPLE REPLICATING SIGNALS USING ADAPTIVE FILTERING PROCEDURES.

Area of Science:

• Genetics
• Biostatistics
• Bioinformatics

Background:

• Replicability is essential for scientific validity, ensuring discoveries are detectable across diverse settings.
• Meta-analysis addresses variability but doesn't guarantee replicability; partial conjunction (PC) null testing identifies signals consistently found in multiple studies.
• Simultaneous testing of numerous PC nulls in high-throughput experiments necessitates robust multiple comparisons correction.

Purpose of the Study:

• To develop a new multiple testing procedure, AdaFilter, to increase statistical power in identifying replicable scientific signals.
• To address the conservativeness of standard multiple testing adjustments when dealing with a large number of PC nulls and sparse signals.
• To provide a method that controls the family-wise error rate (FWER) and false discovery rate (FDR) under data independence.

Main Methods:

• Introduced AdaFilter, a novel multiple testing procedure.
• AdaFilter adaptively filters out unlikely candidates of PC nulls to enhance power.
• Proved AdaFilter's ability to control FWER and FDR under data independence.

Main Results:

• AdaFilter demonstrates significantly higher statistical power compared to existing methods.
• The procedure effectively controls FWER and FDR, ensuring reliable identification of true signals.
• AdaFilter successfully applied to diverse high-throughput datasets including microarray, single-cell RNA sequencing, and GWAS.

Conclusions:

• AdaFilter offers a powerful and reliable approach for multiple testing in scenarios requiring the identification of replicable signals.
• The method is particularly beneficial for large-scale genetic studies where signal sparsity is common.
• AdaFilter enhances the discovery of true biological signals across independent studies, advancing scientific reproducibility.