MoTrPAC ATAC-Seq QC and Analysis Pipeline

This repository provides MoTrPAC-specific supplements to the ENCODE ATAC-seq pipeline. For additional details not directly related to running the ENCODE ATAC-seq pipeline or processing the results, see the most recent version of the MoTrPAC ATAC-seq QC and Analysis Pipeline MOP, available here.

This documentation is intended to help individuals who are preparing ATAC-seq data for submission to the BIC or processing pilot samples with the full pipeline. For simplicity, this documentation explains how to run the full pipeline on a computer compatible with Conda environments. Users working on the cloud or in other environments can follow ENCODE's documentation as necessary. Post-processing scripts are intended to be useful to all users, regardless of environment.

NOTE: MoTrPAC uses ENCODE ATAC-seq pipeline version 1.7.0 for consistency within the consortium and reproducibilty outside of the consortium.

Important references:

• GitHub repository for the ENCODE ATAC-seq pipeline: https://github.com/ENCODE-DCC/atac-seq-pipeline
• ENCODE ATAC-seq pipeline documentation: https://www.encodeproject.org/atac-seq/
• ENCODE data quality standards: https://www.encodeproject.org/atac-seq/#standards
• ENCODE terms and definitions: https://www.encodeproject.org/data-standards/terms/

Table of Contents:

1. Prepare ATAC-seq data for submission to the BIC

   1.1 Clone this repository

   1.2 Generate and format FASTQs

   1.3 Collect additional documents

   1.4 Submit data

   1.5 For GET: Download pipeline outputs FROM BIC

2. Install and test ENCODE ATAC-seq pipeline and dependencies

   2.1 Clone the ENCODE repository

   2.2 Install the Conda environment with all software dependencies

   2.3 Initialize Caper

   2.4 Run a test sample

   2.5 Install genome databases

   2.5.1 Install the hg38 genome database

   2.5.2 Install the custom rn6 genome database

3. Run the ENCODE ATAC-seq pipeline

   3.1 Generate configuration files

   3.2 Run the pipeline

4. Organize outputs

   4.1 Collect important outputs with Croo

   4.2 Generate a spreadsheet of QC metrics for all samples with qc2tsv

5. Flag problematic samples

6. Post-processing scripts

1. Prepare ATAC-seq data for submission to the BIC

1.1 Clone this repository

This documentation will assume you clone it in a folder called ATAC_PIPELINE in your home directory. ~/ATAC_PIPELINE is also the recommended destination folder for when you clone ENCODE's repository later.

cd ~
mkdir ATAC_PIPELINE
cd ATAC_PIPELINE
git clone https://github.com/MoTrPAC/motrpac-atac-seq-pipeline.git

1.2 Generate and format FASTQs

Each GET site (Stanford and MSSM) is responsible for sequencing the library and obtaining the demultiplexed FASTQ files for each sample. If sequencing is performed with NovaSeq, raw data is output as BCL files, which must be demultiplexed and converted to FASTQ files with bcl2fastq (version 2.20.0). bcl2fastq v2.20.0 can be downloaded directly from Illumina here.

Prepare a sample sheet for demultiplexing. Find an example here.

• The sample sheet must not include the Adapter or AdapterRead2 settings. This will prevent bcl2fastq from automatically performing adapter trimming, which provides us with FASTQ files that include the fullest form of the raw data. Adapter trimming is performed downstream
• Sample_Name and Sample_ID should correspond to vial labels; FASTQ files must follow the naming convention ${viallabel}_R?.fastq.gz before submission to the BIC

src/bcl2fastq.sh provides code both to run bcl2fastq and rename files. It can be run as follows:

1. Define the following paths:

• bclfolder: Path to sequencing output directory, e.g /lab/data/NOVASEQ_BATCH1/181205_NB551514_0071_AHFHLGAFXY
• samplesheet: Path to the sample sheet, e.g. ${bclfolder}/SampleSheet.csv
• outdir: Path to root output folder, e.g. /lab/data/NOVASEQ_BATCH1

2. If applicable, load the correct version of bcl2fastq. For example, on Stanford SCG, run module load bcl2fastq2/2.20.0.422.
3. Run src/bcl2fastq.sh:

bash ~/ATAC_PIPELINE/motrpac-atac-seq-pipeline/src/bcl2fastq.sh ${bclfolder} ${samplesheet} ${outdir}

This makes two new directories:

1. ${outdir}/bcl2fastq: Outputs of bcl2fastq
2. ${outdir}/fastq_raw: Merged and re-named FASTQ files, ready for submission to the BIC

Alternatively, run the bcl2fastq command independently, and use your own scripts to merge and rename FASTQ files before submission to the BIC:

bcl2fastq \   
     --sample-sheet /path/to/SampleSheet.csv
     --runfolder-dir $seqDir \
     --output-dir $outDir 

This command will generate two FASTQ files (one for each read in the pair) per sample per lane, e.g. ${viallabel}_L${lane}_R{1,2}_001.fastq.gz.

1.3 Collect additional documents

• Collect the laneBarcode HTML report in ${outdir}/bcl2fastq/Reports/html/*/all/all/all/laneBarcode.html. This report must be included in the BIC data submission,
• Generate sample_metadata_YYYYMMDD.csv. See this table for a list of metrics that must be included in this file.
• Generate file_manifest_YYYYMMDD.csv. See the GET CAS-to-BIC Data Transfer Guidelines for details about the format of this document.

1.4 Submit data

Refer to the GET CAS-to-BIC Data Transfer Guidelines for details about the directory structure for ATAC-seq data submissions. The following files are required:

• file_manifest_YYYYMMDD.csv
• sample_metadata_YYYYMMDD.csv
• readme_YYYYMMDD.txt
• laneBarcode.html
• fastq_raw/*.fastq.gz

1.5 For GET: Download pipeline outputs FROM BIC

After the BIC has finished running the ENCODE ATAC-seq pipeline on a batch of submitted data, use pass_extract_atac_from_gcp.sh to download the important subset of outputs from GCP. Inside the script, change the download_dir and gsurl paths to point to the gsutil source and the local destination, respectively. Then run the script with the number of cores available for parallelization as an argument, e.g.:

bash pass_extract_atac_from_gcp.sh 10

2. Install and test ENCODE ATAC-seq pipeline and dependencies

All steps in this section must only be performed once. After dependencies are installed and genome databases are built, skip to here.

The ENCODE pipeline supports many cloud platforms and cluster engines. It also supports docker, singularity, and Conda to resolve complicated software dependencies for the pipeline. There are special instructions for two major Stanford HPC servers (SCG4 and Sherlock).

While the BIC runs this pipeline on Google Cloud Platform, this documentation is tailored for consortium users who use non-cloud computing environments, including clusters and personal computers. Therefore, this documentation describes the Conda implementation. Refer to ENCODE's documentation for alternatives.

2.1 Clone the ENCODE repository

Clone the v1.7.0 ENCODE repository and this repository in a folder in your home directory:

cd ~/ATAC_PIPELINE
git clone --single-branch --branch v1.7.0 https://github.com/ENCODE-DCC/atac-seq-pipeline.git

IMPORTANT: The following change needs to be made to ~/ATAC_PIPELINE/atac-seq-pipeline/atac.wdl.
At the end of ~/ATAC_PIPELINE/atac-seq-pipeline/atac.wdl, find this block of code:

task raise_exception {
	String msg
	command {
		echo -e "\n* Error: ${msg}\n" >&2
		exit 2
	}
	output {
		String error_msg = '${msg}'
	}
	runtime {
		maxRetries : 0
	}
}

Replace the runtime parameters in the raise_exception task with these:

    runtime {
        maxRetries : 0
        cpu : 1
        memory : '2 GB'
        time : 1
        disks : 'local-disk 10 SSD'
    }

If you do not make this change, you will get the following error when you try to run the pipeline:

Task raise_exception has an invalid runtime attribute memory = !! NOT FOUND !!

* Found failures JSON object.
[
    {
        "causedBy": [
            {
                "causedBy": [],
                "message": "Task raise_exception has an invalid runtime attribute memory = !! NOT FOUND !!"
            },
            {
                "causedBy": [],
                "message": "Task raise_exception has an invalid runtime attribute memory = !! NOT FOUND !!"
            }
        ],
        "message": "Runtime validation failed"
    }
]

2.2 Install the Conda environment with all software dependencies

Install conda by following these instructions. Perform Step 5 in a screen or tmux session, as it can take some time.

2.3 Initialize Caper

Installing the Conda environment also installs Caper. Make sure it works:

conda activate encode-atac-seq-pipeline
caper

If you see an error like caper: command not found, then add the following line to the bottom of ~/.bashrc and re-login.

export PATH=$PATH:~/.local/bin

Choose a platform from the following table and initialize Caper. This will create a default Caper configuration file ~/.caper/default.conf, which have only required parameters for each platform. There are special platforms for Stanford Sherlock/SCG users.

$ caper init [PLATFORM]

Platform	Description
sherlock	Stanford Sherlock cluster (SLURM)
scg	Stanford SCG cluster (SLURM)
gcp	Google Cloud Platform
aws	Amazon Web Service
local	General local computer
sge	HPC with Sun GridEngine cluster engine
pbs	HPC with PBS cluster engine
slurm	HPC with SLURM cluster engine

Edit ~/.caper/default.conf according to your chosen platform. Find instruction for each item in the following table.

IMPORTANT: ONCE YOU HAVE INITIALIZED THE CONFIGURATION FILE ~/.caper/default.conf WITH YOUR CHOSEN PLATFORM, THEN IT WILL HAVE ONLY REQUIRED PARAMETERS FOR THE CHOSEN PLATFORM. DO NOT LEAVE ANY PARAMETERS UNDEFINED OR CAPER WILL NOT WORK CORRECTLY.

Parameter	Description
tmp-dir	IMPORTANT: A directory to store all cached files for inter-storage file transfer. DO NOT USE /tmp.
slurm-partition	SLURM partition. Define only if required by a cluster. You must define it for Stanford Sherlock.
slurm-account	SLURM partition. Define only if required by a cluster. You must define it for Stanford SCG.
sge-pe	Parallel environment of SGE. Find one with $ qconf -spl or ask you admin to add one if not exists.
aws-batch-arn	ARN for AWS Batch.
aws-region	AWS region (e.g. us-west-1)
out-s3-bucket	Output bucket path for AWS. This should start with s3://.
gcp-prj	Google Cloud Platform Project
out-gcs-bucket	Output bucket path for Google Cloud Platform. This should start with gs://.

An important optional parameter is db. If you would like to enable call-catching (i.e. re-use ouputs from previous workflows, which is particularly useful if a workflow fails halfway through a pipeline), add the following lines to ~/.caper/default.conf:

db=file
java-heap-run=4G

2.4 Run a test sample

Follow these platform-specific instructions to run a test sample. Use the following variable assignments:

PIPELINE_CONDA_ENV=encode-atac-seq-pipeline
WDL=~/ATAC_PIPELINE/atac-seq-pipeline/atac.wdl
INPUT_JSON=https://storage.googleapis.com/encode-pipeline-test-samples/encode-atac-seq-pipeline/ENCSR356KRQ_subsampled_caper.json

Note that Caper writes all outputs to the current working directory, so first cd to the desired output directory before using caper run or caper server.

Here is an example of how the test workflow is run on Stanford SCG (SLURM):

conda activate ${PIPELINE_CONDA_ENV}
JOB_NAME=encode_test
sbatch -A ${ACCOUNT} -J ${JOB_NAME} --export=ALL --mem 2G -t 4-0 --wrap "caper run ${WDL} -i ${INPUT_JSON}"

2.5 Install genome databases

2.5.1 Install the hg38 genome database

Specify a destination directory and install the ENCODE hg38 reference with the following command. We recommend not to run this installer on a login node of your cluster. It will take >8GB memory and >2h time.

conda activate encode-atac-seq-pipeline
outdir=/path/to/reference/genome/hg38
bash ~/ATAC_PIPELINE/atac-seq-pipeline/scripts/download_genome_data.sh hg38 ${outdir}  

2.5.2 Install the custom rn6 genome database

Find this section in ~/ATAC_PIPELINE/atac-seq-pipeline/scripts/build_genome_data.sh:

...

elif [[ "${GENOME}" == "YOUR_OWN_GENOME" ]]; then
  # Perl style regular expression to keep regular chromosomes only.
  # this reg-ex will be applied to peaks after blacklist filtering (b-filt) with "grep -P".
  # so that b-filt peak file (.bfilt.*Peak.gz) will only have chromosomes matching with this pattern
  # this reg-ex will work even without a blacklist.
  # you will still be able to find a .bfilt. peak file
  # use ".*", which means ALL CHARACTERS, if you want to keep all chromosomes
  # use "chr[\dXY]+" to allow chr[NUMBERS], chrX and chrY only
  # this is important to make your final output peak file (bigBed) work with genome browsers
  REGEX_BFILT_PEAK_CHR_NAME=".*"
  # REGEX_BFILT_PEAK_CHR_NAME="chr[\dXY]+"

  # mitochondrial chromosome name (e.g. chrM, MT)
  MITO_CHR_NAME="chrM"
  # URL for your reference FASTA (fasta, fasta.gz, fa, fa.gz, 2bit)
  REF_FA="https://some.where.com/your.genome.fa.gz"
  # 3-col blacklist BED file to filter out overlapping peaks from b-filt peak file (.bfilt.*Peak.gz file).
  # leave it empty if you don't have one
  BLACKLIST=
fi
...

Above it, add this block:

elif [[ "${GENOME}" == "motrpac_rn6" ]]; then
  REGEX_BFILT_PEAK_CHR_NAME=".*"
  MITO_CHR_NAME="chrM"
  REF_FA="http://mitra.stanford.edu/montgomery/projects/motrpac/atac/SCG/motrpac_references/rn6_release96/Rattus_norvegicus.Rnor_6.0.dna.toplevel.standardized.fa.gz"
  TSS="http://mitra.stanford.edu/montgomery/projects/motrpac/atac/SCG/motrpac_references/rn6_release96/Rattus_norvegicus.Rnor_6.0.96_protein_coding.tss.bed.gz"
  BLACKLIST=

NOTE: The TSS reference file was generated from the Ensembl GTF using this script.

Now run the script to build the custom genome database. Specify a destination directory and install the MoTrPAC rn6 reference with the following command. We recommend not to run this installer on a login node of your cluster. It will take >8GB memory and >2h time.

conda activate encode-atac-seq-pipeline
outdir=/path/to/reference/genome/motrpac_rn6
bash ~/ATAC_PIPELINE/atac-seq-pipeline/scripts/build_genome_data.sh motrpac_rn6 ${outdir}

3. Run the ENCODE ATAC-seq pipeline

MoTrPAC will run the ENCODE pipeline both with singletons for human samples and replicates for rat samples. In both cases, many iterations of the pipeline will need to be run for each batch of sequencing data. This repository provides scripts to automate this process, for both rat and human samples.

Running the pipeline with replicates outputs all of the same per-sample information generated by running the pipeline with a single sample but improves power for peak calling and outputs a higher-confidence peak set called using all replicates. This generates a single peak set for every exercise protocol/timepoint/tissue/sex combination in the PASS study, which will be useful for downstream analyses.

3.1 Generate configuration files

A configuration (config) file in JSON format that specifies input parameters is required to run the pipeline. Find comprehensive documentation of definable parameters here.

Please click the appropriate link below for detailed instructions on how to automate the generation of config files for pipelines with singletons or replicates. This is particularly important for PASS data, as this repository provides a script to automatically group replicates in the same condition (protocol/timepoint/tissue/sex).

• Prepare config files for replicates (PASS/rat)
• Prepare config files for singletons (CASS/human)

3.2 Run the pipeline

Actually running the pipeline is straightforward. However, the command is different depending on the environment in which you set up the pipeline. Refer back to environment-specific instructions here.

An atac directory containing all of the pipeline outputs is created in the output directory (note the default output directory is the current working directory). One arbitrarily-named subdirectory for each config file (assuming the command is run in a loop for several samples) is written in atac.

Here is an example of code that submits a batch of pipelines to the Stanford SCG job queue. ${JSON_DIR} is the path to all of the config files, generated in Step 3.1:

conda activate encode-atac-seq-pipeline

ATACSRC=~/ATAC_PIPELINE/atac-seq-pipeline
OUTDIR=/path/to/output/directory
cd ${OUTDIR}

for json in $(ls ${JSON_DIR}); do 
  
  INPUT_JSON=${JSON_DIR}/${json}
  JOB_NAME=$(basename ${INPUT_JSON} | sed "s/\.json.*//")

  sbatch -A ${ACCOUNT} -J ${JOB_NAME} --export=ALL --mem 2G -t 4-0 --wrap "caper run ${ATACSRC}/atac.wdl -i ${INPUT_JSON}"
  sleep 60 # necessary to prevent a collision error
done

4. Organize outputs

4.1 Collect important outputs with Croo

Croo is a tool ENCODE developed to simplify the pipeline outputs. It was installed along with the Conda environment. Run it on each sample in the batch. See Table 4.1 for a description of outputs generated by this process.

conda activate encode-atac-seq-pipeline

cd ${OUTDIR}/atac
for dir in $(ls */metadata.json | sed "s:/metadata\.json::"); do 
  cd $dir
  croo metadata.json 
  cd ..
done

Table 4.1. Important files in Croo-organized ENCODE ATAC-seq pipeline output.

Subdirectory or file	Description
qc/*	Components of the merged QC spreadhseet (see Step 4.2)
signal/*/*fc.signal.bigwig	MACS2 peak-calling signal (fold-change), useful for visualizing "read pileups" in a genome browser
signal/*/*pval.signal.bigwig	MACS2 peak-calling signal (P-value), useful for visualizing "read pileups" in a genome browser. P-value track is more dramatic than the fold-change track
align/*/*.trim.merged.bam	Unfiltered BAM files
align/*/*.trim.merged.nodup.no_chrM_MT.bam	Filtered BAM files, used as input for peak calling
align/*/*.tagAlign.gz	tagAlign files from filtered BAMs
peak/overlap_reproducibility/ overlap.optimal_peak.narrowPeak.hammock.gz	Hammock file of overlap peaks, optimized for viewing peaks in a genome browser
peak/overlap_reproducibility/ overlap.optimal_peak.narrowPeak.gz	BED file of overlap peaks. Generally, use this as your final peak set
peak/overlap_reproducibility/ overlap.optimal_peak.narrowPeak.bb	bigBed file of overlap peaks useful for visualizing peaks in a genome browser
peak/idr_reproducibility/ idr.optimal_peak.narrowPeak.gz	IDR peaks. More conservative than overlap peaks

ENCODE recommends using the overlap peak sets when one prefers a low false negative rate but potentially higher false positives; they recommend using the IDR peaks when one prefers low false positive rates.

4.2 Generate a spreadsheet of QC metrics for all samples with qc2tsv

This is most useful if you ran the pipeline for multiple samples. Step 4.1 generates a qc/qc.json file for each pipeline run. After installing qc2tsv within the encode-atac-seq-pipeline Conda environment (pip install qc2tsv), run the following command to compile a spreadsheet with QC from all samples:

cd ${outdir}/atac
qc2tsv $(find -path "*/call-qc_report/execution/glob-*/qc.json") --collapse-header > spreadsheet.tsv

Table 4.2 provides definitions for a limited number of metrics included in the JSON QC reports. The full JSON report includes >100 metrics per sample; some lines are duplicates, and many metrics are irrelevant for running the pipeline with a single biological replicate.

Table 4.2. Description of relevant QC metrics.

Metric	Definition/Notes
replication.reproducibility.overlap.N_opt	Number of optimal overlap_reproducibility peaks
replication.reproducibility.overlap.opt_set	Peak set corresponding to optimal overlap_reproducibility peaks
replication.reproducibility.idr.N_opt	Number of optimal idr_reproducibility peaks
replication.reproducibility.idr.opt_set	Peak set corresponding to optimal idr_reproducibility peaks
replication.num_peaks.num_peaks	Number of peaks called in each replicate
peak_enrich.frac_reads_in_peaks.macs2.frip	Replicate-level FRiP in raw MACS2 peaks
peak_enrich.frac_reads_in_peaks.overlap.{opt_set}.frip	Many FRiP values are reported. In order to get the FRiP corresponding to the overlap_reproducibility peak set, you need to cross-reference the replication.reproducibility.overlap.opt_set metric with these column names to extract the appropriate FRiP. For example, if replication.reproducibility.overlap.opt_set is pooled-pr1_vs_pooled-pr2, then you need to extract the FRiP value from the peak_enrich.frac_reads_in_peaks.overlap.pooled-pr1_vs_pooled-pr2.frip column. See insert script name to see how to do this in an automated way
peak_enrich.frac_reads_in_peaks.idr.{opt_set}.frip	Cross-reference with replication.reproducibility.idr.opt_set. See peak_enrich.frac_reads_in_peaks.overlap.{opt_set}.frip
align.samstat.total_reads	Total number of alignments* (including multimappers)
align.samstat.pct_mapped_reads	Percent of reads that mapped
align.samstat.pct_properly_paired_reads	Percent of reads that are properly paired
align.dup.pct_duplicate_reads	Fraction (not percent) of read pairs that are duplicates after filtering alignments for quality
align.frac_mito.frac_mito_reads	Fraction of reads that align to chrM after filtering alignments for quality and removing duplicates
align.nodup_samstat.total_reads	Number of alignments* after applying all filters
align.frag_len_stat.frac_reads_in_nfr	Fraction of reads in nucleosome-free-region. Should be a value greater than 0.4
align.frag_len_stat.nfr_over_mono_nuc_reads	Reads in nucleosome-free-region versus reads in mononucleosomal peak. Should be a value greater than 2.5
align.frag_len_stat.nfr_peak_exists	Does a nucleosome-free-peak exist? Should be true
align.frag_len_stat.mono_nuc_peak_exists	Does a mononucleosomal-peak exist? Should be true
align.frag_len_stat.di_nuc_peak_exists	Does a dinucleosomal-peak exist? Ideally true, but not condemnable if false
lib_complexity.lib_complexity.NRF	Non-reduandant fraction. Measure of library complexity, i.e. degree of duplicates. Ideally >0.9
lib_complexity.lib_complexity.PBC1	PCR bottlenecking coefficient 1. Measure of library complexity. Ideally >0.9
lib_complexity.lib_complexity.PBC2	PCR bottlenecking coefficient 2. Measure of library complexity. Ideally >3
align_enrich.tss_enrich.tss_enrich	TSS enrichment

*Note: Alignments are per read, so for PE reads, there are two alignments per fragment if each PE read aligns once.

5. Flag problematic samples

The following metrics are not strictly exclusion criteria for MoTrPAC samples, but samples should be flagged if any of these conditions are met. Some of these metrics reflect the ENCODE ATAC-seq data standards.

Table 5.1 Criteria to flag problematic samples.

Description	In terms of Table 2 metrics	Comments
< 50 million filtered, non-duplicated, non-mitochondrial paired-end reads in the filtered BAM file (i.e. 25M pairs)	align.nodup_samstat.total_reads < 50M	This is the most stringent criterion and may be relaxed
Alignment rate < 80%	align.samstat.pct_mapped_reads < 80%
Fraction of reads in overlap peaks < 0.1	peak_enrich.frac_reads_in_peaks.overlap.{opt_set}.frip < 0.1	This is more relaxed than the ENCODE recommendation. Note that replicate-level FRiP in raw peaks can be assessed with peak_enrich.frac_reads_in_peaks.macs2.frip
Number of peaks in overlap peak set < 80,000	replication.reproducibility.overlap.N_opt < 80000	This is more relaxed than the ENCODE recommendation
A nucleosome-free region is not present	align.frag_len_stat.nfr_peak_exists == false	This should be enforced more strictly
A mononucleosome peak is not present	align.frag_len_stat.mono_nuc_peak_exists == false	This should be enforced more strictly
TSS enrichment < ?	align_enrich.tss_enrich.tss_enrich	This cutoff needs to be evaluated retrospectively. We will probably have tissue-specific recommendations

6. Post-processing scripts

• extract_rep_names_from_encode.sh: generate rep-to-viallabel map to interpret QC report
• pass_extract_atac_from_gcp.sh: download relevant files from ENCODE output
• encode_to_count_matrix.sh: use narrowkpeak.gz and tagAlign files to generate a peak x sample raw counts matrix
• align_stats.sh: calculate % of primary alignments aligning to chrX, chrY, chrM, autosomes, and contigs
• merge_atac_qc.R: merge wet lab QC, curated pipeline QC, and alignment stats