Fig. 1

High-throughput omics data analyses and generic FDR control methods. (a) Illustration of four common high-throughput omics data analyses: peak calling from ChIP-seq data, peptide identification from MS data, DEG analysis from RNA-seq data, and DIR analysis from Hi-C data. In these four analyses, the corresponding features are genomic regions (yellow intervals), peptide-spectrum matches (PSMs; a pair of a mass spectrum and a peptide sequence), genes (columns in the heatmaps), and chromatin interacting regions (entries in the heatmaps). (b) Illustration of Clipper and five generic FDR control methods: BH-pair (and qvalue-pair), BH-pool (and qvalue-pool), and locfdr. The input data are d features with m and n repeated measurements under the experimental and background conditions, respectively. Clipper computes a contrast score for each feature based on the feature’s m and n measurements, decides a contrast-score cutoff, and calls the features with contrast scores above the cutoff as discoveries. (This illustration is Clipper for enrichment analysis with m=n.) BH-pair or qvalue-pair computes a p-value for each feature based on the feature’s m and n measurements, sets a p-value cutoff, and calls the features with p-values below the cutoff as discoveries. BH-pool or qvalue-pool constructs a null distribution from the d features’ average (across the n replicates) measurements under the background condition, calculates a p-value for each feature based on the null distribution and the feature’s average (across the m replicates) measurements under the experimental condition, sets a p-value cutoff, and calls the features with p-values below the cutoff as discoveries. The locfdr method computes a summary statistic for each feature based on the feature’s m and n measurements, estimates the empirical null distribution and the empirical distribution of the statistic across features, computes a local fdr for each feature, sets a local fdr cutoff, and calls the features with local fdr below the cutoff as discoveries