Fig. 4
From: STARRPeaker: uniform processing and accurate identification of STARR-seq active regions
Overview of STARRPeaker peak-calling scheme. a In contrast to using read depth (gray), fragment depth (red) offers more precise and sharper STARR-seq output coverage. Fragment inserts are directly inferred from properly paired reads. b Workflow of STARRPeaker describing how coverage is calculated for each genomic bin and modeled using a negative binomial regression model. The analysis pipeline can largely be divided into four steps: (1) binning the genome; (2) calculating coverage and computing covariate matrix; (3) fitting the STARR-seq data to the NB regression model; and (4) peak calling, multiple hypothesis testing correction, and adjustment of the center of peaks