A non-randomized procedure for large-scale heterogeneous multiple discrete testing based on randomized tests.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Next-generation sequencing generates massive discrete data, posing challenges for traditional statistical methods.
• Existing multiple testing procedures often assume continuous data, leading to conservative results for discrete genomic tests.

Purpose of the Study:

• To develop a novel multiple testing procedure for improved false discovery rate (FDR) control on heterogeneous discrete genomic tests.
• To overcome the conservativeness of existing methods in analyzing discrete sequencing data.

Main Methods:

• Proposed a novel multiple testing procedure based on the marginal critical function (MCF) of randomized tests.
• Derived upper bounds for positive FDR (pFDR) and positive false non-discovery rate (pFNR).
• Demonstrated method superiority through simulations and analysis of whole-genome bisulfite sequencing (WGBS) data for differentially methylated region (DMR) detection.

Main Results:

• The proposed procedure achieves powerful and non-randomized multiple testing for discrete data.
• Guaranteed that the method's detections encompass those from naive q-value applications.
• Showcased improved performance over existing methods in simulations and real WGBS data analysis.

Conclusions:

• The novel procedure offers a more powerful and accurate approach to FDR control in heterogeneous discrete testing scenarios common in genomics.
• This method enhances the analysis of next-generation sequencing data, particularly for identifying biological regions like DMRs.