REFERENCE

November 8, 2021 · View on GitHub

This document contains more detailed information on the inputs, outputs and the software.

Software
Genome Reference Files
Inputs
Resource Considerations
Outputs
pBAM

Software

Ubuntu 16.04

The pipeline docker image is based on Ubuntu base image version 16.04.

Python 3.5.2

Python parts of the pipeline are run using Python 3.5.2 that ships with Ubuntu 16.04.

R 3.2.3

MAD QC is run using R version 3.2.3 (2015-12-10) -- "Wooden Christmas-Tree".

STAR 2.5.1b

Alignment is done using STAR 2.5.1b. For detailed description of the software see Article by Dobin et al. Multiple versions of the software have been released since writing the article.

Quantification is done using RSEM v1.2.31. For detailed description of the software see Article by Bo Li and Colin N Dewey. Multiple versions of the software have been released since writing the article.

Kallisto 0.44.0

Additional quantification is calculated using Kallisto v0.44.0. For detailed description of the software see Article by Bray et al..

Samtools 1.9

Samtools flagstats are calculated using Samtools 1.9. For original publication describing the software, see Article by Li H et al. Multiple versions of the software have been released since writing the article.

bedGraphToBigWig and bedSort

bedGraphToBigWig kent source version 371 and bedSort (no version information available) are used in signal track generation. Source code is downloadable here.

Genome Reference Files

Reference and index files can be downloaded from the ENCODE Portal. These files are for human, and mouse will follow. Files that are needed are STAR Index, RSEM Index, Kallisto Index, chromosome sizes file. Links to required files are summarized in the table below.

Genome	Annotation Version	STAR	RSEM	Kallisto	Chr Sizes
GRCh38	Gencode V29	GRCh38 V29 STAR index	GRCh38 V29 RSEM index	V29 Kallisto index	GRCh38 Chr sizes
GRCh38	Gencode V24	GRCh38 V24 STAR index	GRCh38 V24 RSEM index	V24 Kallisto index	GRCh38 Chr sizes
mm10	Gencode M21	mm10 M21 STAR index	mm10 M21 RSEM index	M21 Kallisto index	mm10 Chr sizes

Inputs

Input is given in a json file like this (note that the resources in this input file are suitable for small subsampled test files, for recommendations that work for typical full sized files, see Resource Considerations):

{
    "rna.endedness" : "paired",
    "rna.fastqs_R1" : [["test_data/ENCSR653DFZ_rep1_chr19_10000reads_R1.fastq.gz"], ["test_data/ENCSR653DFZ_rep2_chr19_10000reads_R1.fastq.gz"]],
    "rna.fastqs_R2" : [["test_data/ENCSR653DFZ_rep1_chr19_10000reads_R2.fastq.gz"], ["test_data/ENCSR653DFZ_rep2_chr19_10000reads_R2.fastq.gz"]],
    "rna.align_index" : "test_data/GRCh38_v24_ERCC_phiX_starIndex_chr19only.tgz",
    "rna.rsem_index" : "test_data/GRCh38_v24_ERCC_phiX_rsemIndex_chr19only.tgz",
    "rna.bamroot" : "PE_stranded",
    "rna.strandedness" : "stranded",
    "rna.strandedness_direction" : "reverse",
    "rna.chrom_sizes" : "test_data/GRCh38_EBV.chrom.sizes",
    "rna.align_ncpus" : 2,
    "rna.align_ramGB" : 4,
    "rna.kallisto_index" : "test_data/Homo_sapiens.GRCh38.cdna.all.chr19_ERCC_phix_k31_kallisto.idx",
    "rna.kallisto_number_of_threads" : 2,
    "rna.kallisto_ramGB" : 4,
    "rna.rna_qc_tr_id_to_gene_type_tsv" : "transcript_id_to_gene_type_mappings/gencodeV24pri-tRNAs-ERCC-phiX.transcript_id_to_genes.tsv",
    "rna.bam_to_signals_ncpus" : 1,
    "rna.bam_to_signals_ramGB" : 2,
    "rna.rsem_ncpus" : 2,
    "rna.rsem_ramGB" : 4,
    "rna.align_disk" : "local-disk 20 HDD",
    "rna.kallisto_disk" : "local-disk 20 HDD",
    "rna.rna_qc_disk" : "local-disk 20 HDD",
    "rna.mad_qc_disk" : "local-disk 20 HDD",
    "rna.bam_to_signals_disk" : "local-disk 20 HDD",
    "rna.rsem_disk" : "local-disk 20 HDD"
}

Following elaborates the meaning of each line in the input file.

rna.endedness Indicates whether the endedness of the experiment is paired or single.
rna.fastqs_R1 Is list of lists of gzipped fastq files containing the first pairs of reads.
rna.fastqs_R2 Is list of lists of gzipped fastq files containing the second pairs of reads.

Example:

Assume you are running a paired end experiment with 3 replicates. The fastq files from the first replicate are replicate1_read1.fastq.gz and replicate1_read2.fastq.gz. The fastq files from the second replicate are replicate2_read1.fastq.gz and replicate2_read2.fastq.gz. Finally assume that the fastq files from the third replicate are replicate3_read1.fastq.gz and replicate3_read2.fastq.gz. In this case the input on the relevant part should be as follows: "rna.fastqs_R1" : [["replicate1_read1.fastq.gz"], ["replicate2_read1.fastq.gz"], ["replicate3_read1.fastq.gz"]] "rna.fastqs_R2" : [["replicate1_read2.fastq.gz"], ["replicate2_read2.fastq.gz"], ["replicate3_read2.fastq.gz"]] Note that it is very important that the replicates are in same order in both lists, this correspondence is used for pairing correct files with each other.

Example:

Assume you are running a single end experiment with 3 replicates. Further assume that in your first round of sequencing you do not get depth you would like (but you still think that the data you obtained is good), and that the data for the replicates are in files replicate1_part1.fastq.gz and replicate2_part1.fastq.gz. To obtain a deeper data, you submit the libraries from the replicates to re-sequencing, which yields data files replicate1_part2.fastq.gz and replicate2_part2.fastq.gz. In this case the input on the fastq part should be: "rna.fastqs_R1" : [["replicate1_part1.fastq.gz", "replicate1_part2.fastq.gz"], ["replicate2_part1.fastq.gz", "replicate2_part2.fastq.gz"]] "rna.fastqs_R2" : [] (in single ended experiment the second read input is empty) In the case when the input for a replicate consists of several fastq files, they will be automatically merged during the pipeline run.

rna.align_index Is the index for STAR aligner.
rna.rsem_index Is the index for RSEM quantifier.
rna.kallisto_index Is the index for Kallisto quantifier. Required only if rna.run_kallisto is true (Default is true).
rna.bamroot This is a prefix that gets added into the output filenames. Additionally the files are prefixed with information of the replicate they originate from.

Example:

Assume the rna.bamroot is FOO. Outputs from first replicate would be prefixed by rep1FOO and outputs from second replicate would be prefixed by rep2FOO etc.

rna.strandedness Indicates whether the experiment is stranded or unstranded. If this is stranded, then the rna.strandedness_direction should be set to forward or reverse.
rna.strandedness_direction Indicates the direction of strandedness. Options are forward, reverse and unstranded.
rna.chrom_sizes Is the file containing the chromosome sizes. You can find and download the files from ENCODE Portal, see table above.
rna.align_ncpus How many cpus are available for STAR alignment.
rna.align_ramGB How many GBs of memory are available for STAR alignment.
rna.align_ncpus How many cpus are available for RSEM quantification.
rna.align_ramGB How many GBs of memory are available for RSEM quantification.
rna.align_disk How much disk space is available for Align task. You can also specify the type of disk, HDD for a spinning disk and SSD for a solid state drive.
rna.kallisto_disk As above, but for Kallisto. Required only if rna.run_kallisto is true (Default is true).
rna.rna_qc_disk As above, but for RNA QC.
rna.bam_to_signals_disk As above, but for bam_to_signals.
rna.mad_qc_disk As above, but for MAD QC.
rna.rsem_disk As above, but for RSEM.
rna.kallisto_number_of_threads How many threads are used for Kallisto quantification. Required only if rna.run_kallisto is true (Default is true).
rna.kallisto_ramGB How many GBs of memory are available for Kallisto quantification. Required only if rna.run_kallisto is true (Default is true).
rna.docker and rna.singularity Container image locations. You should usually not touch these, unless you are for some reason using locally stored images, in which case you probably know what you are doing :)

Example:

Assume you want to allocate 100 gigabytes of spinning hard drive. In this case you would enter "local-disk 100 HDD". If you want to allocate 111 gigabytes of solid state drive space, enter "local-disk 111 SSD".

rna.rna_qc_tr_id_to_gene_type_tsv rna_qc task calculates the number of reads by gene type. For this a tsv file that contains a mapping from transcript IDs to gene types is provided. For GRCh38, hg19, and mm10 with ERCC (ambion 1) and PhiX spikes the tsv is provided in this repo. If you are using some other annotation, you can use code here to build your own.
rna.bam_to_signals_ncpus Is the number of cpus given to bam_to_signals task.
rna.bam_to_signals_ramGB Is the amount of memory in GB given to bam_to_signals task.

Additional inputs when running single-ended experiments:

Kallisto quantifier makes use of average fragment lenghts and standard deviations of those lengths. In the case of paired end experiments, those values can be calculated from the data, but in case of single-ended experiment those values must be provided.

rna.kallisto_fragment_length Is an array of average fragment lengths as integers, one per replicate. Required only if rna.run_kallisto is true (Default is true). .
rna.kallisto_sd_of_fragment_length Is and array standard deviations of the fragment lengths as floats, one per replicate. Required only if rna.run_kallisto is true (Default is true).

If you do not have this data available, or if you for some other reason want to omit running kallisto you can use the following parameter:

rna.run_kallisto Boolean defaulting to true. If set to false kallisto will not be run, and you do not need to provide values for any kallisto related parameters.

pBAM

This pipeline offers an option to also output the alignments in privacy-preserving BAM, also known as pBAM format. For more details about the format, see Cell article by Gursoy et al.. Note, that also the usual BAM is output, and making sure only the privacy-preserving BAM is published is up to you. All the downstream operations (RSEM quantification, signal track generation, MAD QC and RNA QC) are based on pBAMs. Note that kallisto step, that is directly based on fastqs will not be changed when using this option, and kallisto quantifications might leak some information. Safest way to handle possible leakage from kallisto is to just skip it by setting rna.run_kallisto to false.

Enabling the pBAM option

To enable pBAM producing the following additional inputs are required:

rna.produce_pbams needs to be set to true.
rna.reference_genome The reference genome fasta compressed with gzip. The human reference genome used by ENCODE can be found here.
rna.reference_transcriptome The reference transcriptome fasta compressed with gzip. The version should match the genome annotation version. The human version 29 transcriptome corresponding to ENCODE4 data can be found here.
rna.reference_annotations An array of annotation gtf files compressed with gzip. Typically this would be the genome annotation and tRNA annotation.

Outputs

Cromwell: Cromwell will store outputs for each task under directory cromwell-executions/[WORKFLOW_ID]/call-[TASK_NAME]/shard-[IDX]. For all tasks [IDX] means a zero-based index for each replicate. In addition to the actual pipeline outputs, these directories contain a plethora of operational Cromwell-specific files. Use croo to find and organize the pipeline outputs.

Output files

Task Align

genomebam, file name matches *_genome.bam. Bam aligned to genome.
annobam, file name matches *_anno.bam. Bam aligned to annotation.
genome_flagstat file name matches *_genome_flagstat.txt. Samtools flagstats on the genome bam.
genome_flagstat_json file name matches *_genome_flagstat.json. Samtools flagstats on the genome bam in json format.
anno_flagstat file name matches *_anno_flagstat.txt. Samtools flagstats on anno bam.
anno_flagstat_json file name matches *_anno_flagstat.json. Samtools flagstats on anno bam in json format.
log file name matches *_Log.final.out. STAR run log.
log_json file name matches *_Log.final.out. STAR run log in json format.
python_log file name is align.log. This file contains possible additional information on the pipeline step.

Task Kallisto

Kallisto related outputs will be produced only, if the rna.run_kallisto is set to true (Default is true).

quants, file name matches *_abundance.tsv. Kallisto quantifications.
python_log file name is kallisto_quant.log. This file contains possible additional information on the pipeline step.

Task Bam to Signals

In case of an stranded run, the plus and minus strand signal tracks are separated (there will be four tracks per replicate).

unique_unstranded, file name matches *genome_uniq.bw. Contains the signal track of the uniquely mapped reads.
all_unstranded, the file name matches *genome_all.bw. Contains the signal track of all reads.
unique_plus, file name matches *genome_plusUniq.bw. Contains the signal track of the uniquely mapped reads from plus strand.
all_plus, the file name matches *genome_plusAll.bw. Contains the signal track of all reads from plus strand.
unique_minus, file name matches *genome_minusUniq.bw. Contains the signal track of the uniquely mapped reads from minus strand.
all_minus, the file name matches *genome_minusAll.bw. Contains the signal track of all reads from minus strand.
python_log file name is bam_to_signals.log. This file contains possible additional information on the pipeline step.

Task RSEM Quant

genes_results, file name matches *.genes.results. Contains gene quantifications.
isoforms_results, file name matches *.isoforms.results. Contains isoform quantifications.
number_of_genes, file name matches *_number_of_genes_detected.json. Contains the number of genes detected, which is determined as TPM value being greater than 1.
python_log file name is rsem_quant.log. This file contains possible additional information on the pipeline step.

Task Mad QC

This step is run if and only if the number of replicates is 2.

madQCplot, file name matches *_mad_plot.png. Contains the MAD QC plot.
madQCmetrics file name matches *_mad_qc_metrics.json. Contains MAD QC metrics.
python_log file name is mad_qc.log. This file contains possible additional information on the pipeline step.

Task RNA QC

This step calculates additional metrics. At this time the only metric is to calculate reads by gene type. It is very IMPORTANT to look at the python_log of this step to see that the transcriptome bam did not contain any transcripts that are not present in the transcript ID to gene type mapping tsv. In case that happens, make sure you are using the STAR aligner and RSEM quantifier indexes you think you are using, and that all the other references are correct!

rnaQC, file name matches *_qc.json. Contains additional QC metrics. For now the reads by gene type.
python_log file name is rna_qc.log. This file contains IMPORTANT information on the pipeline step.

Note about resources:

The hardware resources needed to run the pipeline depend on the sequencing depth so it is hard to give definitive values that will be good for everyone. However, for every pipeline run, alignment is going to be the most memory-intensive task, quantitation with RSEM is going to be computationally hardest, kallisto will require some non-trivial amount of resources, and typically the rest of the tasks are rather light both in CPU and memory use. Disk usage again depends on the sequencing depth, but "local-disk 100 HDD" is a good starting point for all the tasks. The following are recommendations that are a sensible starting point for further optimizations in a typical case (non-CPU or memory related inputs omitted):

{
    "rna.align_ncpus" : 16,
    "rna.align_ramGB" : 60,
    "rna.rsem_ncpus" : 16,
    "rna.rsem_ramGB" : 60,
    "rna.kallisto_number_of_threads" : 8,
    "rna.kallisto_ramGB" : 30,
    "rna.bam_to_signals_ncpus" : 8,
    "rna.bam_to_signals_ramGB" : 30
}

In case of building index files for STAR and RSEM the sufficient amount of memory for GRCh38 is 60GB. The merge annotation workflow with gencode V29 annotation and tRNAs works with the default resources defined in the merge_anno.wdl. Smaller references may be able to run with less, but because index building typically needs to be done only once, it is most prudent to overshoot rather than waste time on several attempts. Following inputs with non-subsampled, full sized inputs have been tested on Google Cloud are good landmarks for defining the resources on other platforms as well:

Merge Annotation:

{
    "merge_anno.annotation" : "https://www.encodeproject.org/files/gencode.v29.primary_assembly.annotation_UCSC_names/@@download/gencode.v29.primary_assembly.annotation_UCSC_names.gtf.gz",
    "merge_anno.tRNA" : "https://www.encodeproject.org/files/gencode.v29.tRNAs/@@download/gencode.v29.tRNAs.gtf.gz",
    "merge_anno.spikeins" : "https://github.com/ENCODE-DCC/rna-seq-pipeline/raw/master/test_data/ERCC_phiX.fa.gz",
    "merge_anno.output_filename" : "merged_annotation_V29.gtf.gz"
}

STAR index:

{
  "build_index.reference_sequence" : "GRCh38_no_alt_analysis_set_GCA_000001405.15.fasta.gz",
  "build_index.spikeins" : "ERCC_phiX.fa.gz",
  "build_index.annotation" : "merged_annotation_V29.gtf.gz",
  "build_index.anno_version" : "v29",
  "build_index.genome" : "GRCh38",
  "build_index.index_type" : "prep_star",
  "build_index.ncpu" : 16,
  "build_index.memGB" : 60
}

RSEM index:

{
  "build_index.reference_sequence" : "GRCh38_no_alt_analysis_set_GCA_000001405.15.fasta.gz",
  "build_index.spikeins" : "ERCC_phiX.fa.gz",
  "build_index.annotation" : "merged_annotation_V29.gtf.gz",
  "build_index.anno_version" : "v29",
  "build_index.genome" : "GRCh38",
  "build_index.index_type" : "prep_rsem",
  "build_index.ncpu" : 16,
  "build_index.memGB" : 60
}

Kallisto index:

{
    "build_index.reference_sequence" : "gencode.v29.transcripts_ERCC_phiX.fa.gz",
    "build_index.index_type" : "prep_kallisto",
    "build_index.ncpu" : 16,
    "build_index.memGB" : 60
}

Main pipeline:

{
  "rna.endedness": "single",
  "rna.fastqs_R1": [
    [
      "https://www.encodeproject.org/files/ENCFF824LLV/@@download/ENCFF824LLV.fastq.gz",
      "https://www.encodeproject.org/files/ENCFF729YAX/@@download/ENCFF729YAX.fastq.gz"
    ],
    [
      "https://www.encodeproject.org/files/ENCFF481BWJ/@@download/ENCFF481BWJ.fastq.gz",
      "https://www.encodeproject.org/files/ENCFF974EKR/@@download/ENCFF974EKR.fastq.gz"
    ]
  ],
  "rna.bamroot": "ENCSR843RJV_testing",
  "rna.align_index": "gs://long_read_rna_production/caper_out/build_index/9320f158-6b7f-400a-899e-5b3f7e2ca46e/call-make_index/glob-08af9cb4bb3f985849894cfeafd3be7b/GRCh38_v29_ERCC_phiX_starIndex.tgz",
  "rna.rsem_index": "gs://long_read_rna_production/caper_out/build_index/b15d94e9-b4ef-49dc-b164-45000a0b3cb0/call-make_index/glob-08af9cb4bb3f985849894cfeafd3be7b/GRCh38_v29_ERCC_phiX_rsemIndex.tgz",
  "rna.strandedness": "unstranded",
  "rna.strandedness_direction": "unstranded",
  "rna.chrom_sizes": "gs://release-tests/Inputs/reference/GRCh38_EBV.chrom.sizes",
  "rna.align_ncpus": 16,
  "rna.align_ramGB": 60,
  "rna.kallisto_index": "gs://long_read_rna_production/caper_out/build_index/35ee9e2c-370d-416b-b08d-05307a63b7d1/call-make_index/glob-0a1c08388fcf0d2e32fc9d921d5a8910/gencode.v29.transcripts_ERCC_phiX.idx",
  "rna.kallisto_number_of_threads": 8,
  "rna.kallisto_ramGB": 30,
  "rna.kallisto_fragment_length": [250,250],
  "rna.kallisto_sd_of_fragment_length": [10,10],
  "rna.rna_qc_tr_id_to_gene_type_tsv": "gs://release-tests/Inputs/reference/gencodeV29pri-tRNAs-ERCC-phiX.transcript_id_to_genes.tsv",
  "rna.bam_to_signals_ncpus": 8,
  "rna.bam_to_signals_ramGB": 30,
  "rna.rsem_ncpus": 16,
  "rna.rsem_ramGB": 60,
  "rna.align_disk": "local-disk 200 SSD",
  "rna.kallisto_disk": "local-disk 200 SSD",
  "rna.rna_qc_disk": "local-disk 200 SSD",
  "rna.bam_to_signals_disk": "local-disk 200 SSD",
  "rna.mad_qc_disk": "local-disk 200 SSD",
  "rna.rsem_disk": "local-disk 200 SSD"
}

REFERENCE

CONTENTS

Software

Ubuntu 16.04

Python 3.5.2

R 3.2.3

STAR 2.5.1b

RSEM 1.2.31

Kallisto 0.44.0

Samtools 1.9

bedGraphToBigWig and bedSort

Genome Reference Files

Inputs

Example:

Example:

Example:

Example:

Additional inputs when running single-ended experiments:

pBAM

Enabling the pBAM option

Outputs

Output files

Task Align

Task Kallisto

Task Bam to Signals

Task RSEM Quant

Task Mad QC

Task RNA QC

Note about resources: