![](\"https://s3.amazonaws.com/4dn-dcic-public/static-pages/atac-pipeline.png\")
The 4DN ATAC-seq data processing pipeline uses the ENCODE ATAC-seq pipeline v1.1.1. We have modified the logistics of the pipeline execution without changing the content of the pipeline.
\nWe have split the pipeline into two sub-pipelines; 1) alignment and filtering and 2) peak calling. A quality control report is generated at each step.
\nFor certain cases, we add a replicate-merging step between the two steps (not a part of the ENCODE pipeline).\n* If an experiment set has >1 biological replicates and >1 technical replicates in a biological replicate, we merge the technical replicates.
\nWe perform the basic QC that comes with the ATAC-seq pipeline, but we do not perform ATAQC, an additional QC that requires a set of reference files that do not yet have official release/documentation, i.e. we use the flag atac.disable_ataqc=True.
For more detail, see description/documentation from ENCODE.
\n The first step is run on the fastq files that correspond to a single technical replicate (a single technical replicate may contain multiple sequencing replicates).
\nReads are aligned to the reference genome with bowtie2 and filtered. The output (tagAlign) is a set of read positions in gzipped bed format.
This step is equivalent of running the ENCODE ATAC-seq pipeline with parameter align_only=True.
A quality control report is linked from the main output tagAlign file.
A more detailed description of this step can be found at : Workflow graph and metadata
In some cases, replicates are merged after the first step. The merging rule is as below.
\nA more detailed description of this step can be found at : Workflow graph and metadata
Using TagAlign files obtained from the earlier step, a signal fold change track (in bigwig format) is calculated using MACS2. Peaks are also called using MACS2 and two final call sets (optimal peaks and conservative peaks, in bigbed format) are reported after applying an overlap method. A quality control report is linked from the output signal fold change bigwig file.
A more detailed description of this step can be found at : Workflow graph and metadata
The pipeline components are pre-installed in a publicly available Docker image on Docker Hub (4dn-dcic/encode-chipseq:v2.1.6), which is adapted from the ENCODE docker image (quay.io/encode-dcc/chip-seq-pipeline:v2.1.6). The pipeline structure is described in Workflow Description Language (WDL) and has been modified from the original ENCODE WDL. The source code for the Docker image and the WDL code can be found on GitHub.
Latest runs:
\nThe Docker image for the previous version of the pipeline (4dn-dcic/encode-chipseq:v1.1.1) is identical to the ENCODE docker image (quay.io/encode-dcc/chip-seq-pipeline:v1.1.1). The WDL for this version has also been modified from the original ENCODE WDL.
| File Type | \nFile Format | \nDescription | \n
|---|---|---|
| Read positions | \nbed | \nPositions of aligned reads in bed format. | \n
| Peaks | \nbigBed (.bb) | \nA set of optimal peaks. After peak calling, IDR (irreproducible discovery rate) analysis is performed, which looks for consistency in peak calls. The optimal set consists of the largest set of peaks that pass IDR analysis, derived from either the biological replicates, or from pseudoreplicates (subsampled sets of pooled reads). | \n
| Conservative peaks | \nbigBed (.bb) | \nA set of conservative peaks. Specifically, the set of peak calls that pass IDR analysis of biological replicates. | \n
| Signal fold change | \nbigwig (.bw) | \nFold change in signal. This file can be visualized on the data portal in HiGlass, and has a QC metric associated with it. | \n
| File Type | \nFile Format | \nDescription | \n
|---|---|---|
| Read positions | \nbed | \nPositions of aligned reads in bed format. | \n
| Peaks | \nbigBed (.bb) | \nA set of optimal peaks. After peak calling, IDR (irreproducible discovery rate) analysis is performed, which looks for consistency in peak calls. The optimal set consists of the largest set of peaks that pass IDR analysis, derived from either the biological replicates, or from pseudoreplicates (subsampled sets of pooled reads). | \n
| Conservative peaks | \nbigBed (.bb) | \nA set of conservative peaks. Specifically, the set of peak calls that pass IDR analysis of biological replicates. | \n
| Signal fold change | \nbigwig (.bw) | \nFold change in signal. This file can be visualized on the data portal in HiGlass, and has a QC metric associated with it. | \n