nf-core/rnaseq

Edit

Introduction

This document describes the output produced by the pipeline. Most of the plots are taken from the MultiQC report generated from the full-sized test dataset for the pipeline using a command similar to the one below:

nextflow run nf-core/rnaseq -profile test_full,<docker/singularity/institute>

The directories listed below will be created in the results directory after the pipeline has finished. All paths are relative to the top-level results directory.

Pipeline overview

The pipeline is built using Nextflow and processes data using the following steps:

• nf-core/rnaseq: Output
  • Introduction
  • Pipeline overview
  • Preprocessing
    • cat
    • FastQC
    • UMI-tools extract
    • TrimGalore
    • fastp
    • BBSplit
    • SortMeRNA
  • Alignment and quantification
    • STAR, Salmon and Kallisto
    • STAR via RSEM
    • HISAT2
  • Alignment post-processing
    • SAMtools
    • UMI-tools dedup
    • picard MarkDuplicates
  • Other steps
    • StringTie
    • BEDTools and bedGraphToBigWig
  • Quality control
    • RSeQC
      • Infer experiment
      • Read distribution
      • Junction annotation
      • Inner distance
      • Junction saturation
      • Read duplication
      • BAM stat
      • TIN
    • Qualimap
    • dupRadar
    • Preseq
    • featureCounts
    • DESeq2
    • Kraken2/Bracken
    • MultiQC
  • Pseudoalignment and quantification
    • Pseudoalignment
  • Workflow reporting and genomes
    • Reference genome files
    • Pipeline information

Preprocessing

cat

If multiple libraries/runs have been provided for the same sample in the input samplesheet (e.g. to increase sequencing depth) then these will be merged at the very beginning of the pipeline in order to have consistent sample naming throughout the pipeline. Please refer to the usage documentation to see how to specify these samples in the input samplesheet.

FastQC

FastQC gives general quality metrics about your sequenced reads. It provides information about the quality score distribution across your reads, per base sequence content (%A/T/G/C), adapter contamination and overrepresented sequences. For further reading and documentation see the FastQC help pages.

UMI-tools extract

UMI-tools deduplicates reads based on unique molecular identifiers (UMIs) to address PCR-bias. Firstly, the UMI-tools extract command removes the UMI barcode information from the read sequence and adds it to the read name. Secondly, reads are deduplicated based on UMI identifier after mapping as highlighted in the UMI-tools dedup section.

To facilitate processing of input data which has the UMI barcode already embedded in the read name from the start, --skip_umi_extract can be specified in conjunction with --with_umi.

TrimGalore

Trim Galore! is a wrapper tool around Cutadapt and FastQC to peform quality and adapter trimming on FastQ files. Trim Galore! will automatically detect and trim the appropriate adapter sequence. It is the default trimming tool used by this pipeline, however you can use fastp instead by specifying the --trimmer fastp parameter. You can specify additional options for Trim Galore! via the --extra_trimgalore_args parameters.

NB: TrimGalore! will only run using multiple cores if you are able to use more than > 5 and > 6 CPUs for single- and paired-end data, respectively. The total cores available to TrimGalore! will also be capped at 4 (7 and 8 CPUs in total for single- and paired-end data, respectively) because there is no longer a run-time benefit. See release notes and discussion whilst adding this logic to the nf-core/atacseq pipeline.

fastp

fastp is a tool designed to provide fast, all-in-one preprocessing for FastQ files. It has been developed in C++ with multithreading support to achieve higher performance. fastp can be used in this pipeline for standard adapter trimming and quality filtering by setting the --trimmer fastp parameter. You can specify additional options for fastp via the --extra_fastp_args parameter.

BBSplit

BBSplit is a tool that bins reads by mapping to multiple references simultaneously, using BBMap. The reads go to the bin of the reference they map to best. There are also disambiguation options, such that reads that map to multiple references can be binned with all of them, none of them, one of them, or put in a special “ambiguous” file for each of them.

This functionality would be especially useful, for example, if you have mouse PDX samples that contain a mixture of human and mouse genomic DNA/RNA and you would like to filter out any mouse derived reads.

The BBSplit index will have to be built at least once with this pipeline by providing --bbsplit_fasta_list which has to be a file containing 2 columns: short name and full path to reference genome(s):

mm10,/path/to/mm10.fa
ecoli,/path/to/ecoli.fa
sarscov2,/path/to/sarscov2.fa

You can save the index by using the --save_reference parameter and then provide it via --bbsplit_index for future runs. As described in the Output files dropdown box above the FastQ files relative to the main reference genome will always be called *primary*.fastq.gz.

SortMeRNA

When --remove_ribo_rna is specified, the pipeline uses SortMeRNA for the removal of ribosomal RNA. By default, rRNA databases defined in the SortMeRNA GitHub repo are used. You can see an example in the pipeline Github repository in assets/rrna-default-dbs.txt which is used by default via the --ribo_database_manifest parameter. Please note that commercial/non-academic entities require licensing for SILVA for these default databases.

Alignment and quantification

STAR, Salmon and Kallisto

STAR is a read aligner designed for splice aware mapping typical of RNA sequencing data. STAR stands for Spliced Transcripts Alignment to a Reference, and has been shown to have high accuracy and outperforms other aligners by more than a factor of 50 in mapping speed, but it is memory intensive. Using --aligner star_salmon is the default alignment and quantification option.

The STAR section of the MultiQC report shows a bar plot with alignment rates: good samples should have most reads as Uniquely mapped and few Unmapped reads.

Salmon from Ocean Genomics and Kallisto, from the Pachter Lab, are provided as options for pseudoalignment. Both allow quantification of reads against an index generated from a reference set of target transcripts. By default, the transcriptome-level BAM files generated by STAR are provided to Salmon for downstream quantification, and Kallisto is not an option here (it does not allow BAM file input). But you can provide FASTQ files directly as input to either Salmon or Kallisto in order to pseudoalign and quantify your data by providing the --pseudo_aligner salmon or --pseudo_aligner kallisto parameter. See the Salmon and Kallisto results sections for more details.

STAR via RSEM

RSEM is a software package for estimating gene and isoform expression levels from RNA-seq data. It has been widely touted as one of the most accurate quantification tools for RNA-seq analysis. RSEM wraps other popular tools to map the reads to the genome (i.e. STAR, Bowtie2, HISAT2; STAR is used in this pipeline) which are then subsequently filtered relative to a transcriptome before quantifying at the gene- and isoform-level. Other advantages of using RSEM are that it performs both the alignment and quantification in a single package and its ability to effectively use ambiguously-mapping reads.

You can choose to align and quantify your data with RSEM by providing the --aligner star_rsem parameter.

HISAT2

HISAT2 is a fast and sensitive alignment program for mapping next-generation sequencing reads (both DNA and RNA) to a population of human genomes as well as to a single reference genome. It introduced a new indexing scheme called a Hierarchical Graph FM index (HGFM) which when combined with several alignment strategies, enable rapid and accurate alignment of sequencing reads. The HISAT2 route through the pipeline is a good option if you have memory limitations on your compute. However, quantification isn’t performed if using --aligner hisat2 due to the lack of an appropriate option to calculate accurate expression estimates from HISAT2 derived genomic alignments. However, you can use this route if you have a preference for the alignment, QC and other types of downstream analysis compatible with the output of HISAT2.

You can choose to align your data with HISAT2 by providing the --aligner hisat2 parameter.

Alignment post-processing

The pipeline has been written in a way where all the files generated downstream of the alignment are placed in the same directory as specified by --aligner e.g. if --aligner star_salmon is specified then all the downstream results will be placed in the star_salmon/ directory. This helps with organising the directory structure and more importantly, allows the end-user to get the results from multiple aligners by simply re-running the pipeline with a different --aligner option along the -resume parameter. It also means that results won’t be overwritten when resuming the pipeline and can be used for benchmarking between alignment algorithms if required.

SAMtools

The original BAM files generated by the selected alignment algorithm are further processed with SAMtools to sort them by coordinate, for indexing, as well as to generate read mapping statistics.

UMI-tools dedup

After extracting the UMI information from the read sequence (see UMI-tools extract), the second step in the removal of UMI barcodes involves deduplicating the reads based on both mapping and UMI barcode information using the UMI-tools dedup command. This will generate a filtered BAM file after the removal of PCR duplicates.

picard MarkDuplicates

Unless you are using UMIs it is not possible to establish whether the fragments you have sequenced from your sample were derived via true biological duplication (i.e. sequencing independent template fragments) or as a result of PCR biases introduced during the library preparation. By default, the pipeline uses picard MarkDuplicates to mark the duplicate reads identified amongst the alignments to allow you to guage the overall level of duplication in your samples. However, for RNA-seq data it is not recommended to physically remove duplicate reads from the alignments (unless you are using UMIs) because you expect a significant level of true biological duplication that arises from the same fragments being sequenced from for example highly expressed genes. This step will be skipped automatically when using the --with_umi option or explicitly via the --skip_markduplicates parameter.

Other steps

StringTie

StringTie is a fast and highly efficient assembler of RNA-Seq alignments into potential transcripts. It uses a novel network flow algorithm as well as an optional de novo assembly step to assemble and quantitate full-length transcripts representing multiple splice variants for each gene locus. In order to identify differentially expressed genes between experiments, StringTie’s output can be processed by specialized software like Ballgown, Cuffdiff or other programs (DESeq2, edgeR, etc.).

BEDTools and bedGraphToBigWig

The bigWig format is an indexed binary format useful for displaying dense, continuous data in Genome Browsers such as the UCSC and IGV. This mitigates the need to load the much larger BAM files for data visualisation purposes which will be slower and result in memory issues. The bigWig format is also supported by various bioinformatics software for downstream processing such as meta-profile plotting.

Quality control

RSeQC

RSeQC is a package of scripts designed to evaluate the quality of RNA-seq data. This pipeline runs several, but not all RSeQC scripts. You can tweak the supported scripts you would like to run by adjusting the --rseqc_modules parameter which by default will run all of the following: bam_stat.py, inner_distance.py, infer_experiment.py, junction_annotation.py, junction_saturation.py,read_distribution.py and read_duplication.py.

The majority of RSeQC scripts generate output files which can be plotted and summarised in the MultiQC report.

Infer experiment

This script predicts the “strandedness” of the protocol (i.e. unstranded, sense or antisense) that was used to prepare the sample for sequencing by assessing the orientation in which aligned reads overlay gene features in the reference genome. The strandedness of each sample has to be provided to the pipeline in the input samplesheet (see usage docs). However, this information is not always available, especially for public datasets. As a result, additional features have been incorporated into this pipeline to auto-detect whether you have provided the correct information in the samplesheet, and if this is not the case then the affected libraries will be flagged in the table under ‘Strandedness Checks’ elsewhere in the report. If required, this will allow you to correct the input samplesheet and rerun the pipeline with the accurate strand information. Note, it is important to get this information right because it can affect the final results.

RSeQC documentation: infer_experiment.py

Read distribution

This tool calculates how mapped reads are distributed over genomic features. A good result for a standard RNA-seq experiments is generally to have as many exonic reads as possible (CDS_Exons). A large amount of intronic reads could be indicative of DNA contamination in your sample but may be expected for a total RNA preparation.

RSeQC documentation: read_distribution.py

Junction annotation

Junction annotation compares detected splice junctions to a reference gene model. Splicing annotation is performed in two levels: splice event level and splice junction level.

RSeQC documentation: junction_annotation.py

Inner distance

The inner distance script tries to calculate the inner distance between two paired-end reads. It is the distance between the end of read 1 to the start of read 2, and it is sometimes confused with the insert size (see this blog post for disambiguation):

This plot will not be generated for single-end data. Very short inner distances are often seen in old or degraded samples (eg. FFPE) and values can be negative if the reads overlap consistently.

RSeQC documentation: inner_distance.py

Junction saturation

This script shows the number of splice sites detected within the data at various levels of subsampling. A sample that reaches a plateau before getting to 100% data indicates that all junctions in the library have been detected, and that further sequencing will not yield any more observations. A good sample should approach such a plateau of Known junctions, however, very deep sequencing is typically required to saturate all Novel Junctions in a sample.

RSeQC documentation: junction_saturation.py

Read duplication

This plot shows the number of reads (y-axis) with a given number of exact duplicates (x-axis). Most reads in an RNA-seq library should have a low number of exact duplicates. Samples which have many reads with many duplicates (a large area under the curve) may be suffering excessive technical duplication.

RSeQC documentation: read_duplication.py

BAM stat

This script gives numerous statistics about the aligned BAM files. A typical output looks as follows:

#Output (all numbers are read count)
#==================================================
Total records:                                 41465027
QC failed:                                     0
Optical/PCR duplicate:                         0
Non Primary Hits                               8720455
Unmapped reads:                                0
 
mapq < mapq_cut (non-unique):                  3127757
mapq >= mapq_cut (unique):                     29616815
Read-1:                                        14841738
Read-2:                                        14775077
Reads map to '+':                              14805391
Reads map to '-':                              14811424
Non-splice reads:                              25455360
Splice reads:                                  4161455
Reads mapped in proper pairs:                  21856264
Proper-paired reads map to different chrom:    7648

MultiQC plots each of these statistics in a dot plot. Each sample in the project is a dot - hover to see the sample highlighted across all fields.

RSeQC documentation: bam_stat.py

TIN

This script is designed to evaluate RNA integrity at the transcript level. TIN (transcript integrity number) is named in analogous to RIN (RNA integrity number). RIN (RNA integrity number) is the most widely used metric to evaluate RNA integrity at sample (or transcriptome) level. It is a very useful preventive measure to ensure good RNA quality and robust, reproducible RNA sequencing. This process isn’t run by default - please see this issue.

RSeQC documentation: tin.py

Qualimap

Qualimap is a platform-independent application written in Java and R that provides both a Graphical User Interface (GUI) and a command-line interface to facilitate the quality control of alignment sequencing data. Shortly, Qualimap:

• Examines sequencing alignment data according to the features of the mapped reads and their genomic properties.
• Provides an overall view of the data that helps to to the detect biases in the sequencing and/or mapping of the data and eases decision-making for further analysis.

The Qualimap RNA-seq QC module is used within this pipeline to assess the overall mapping and coverage relative to gene features.

dupRadar

dupRadar is a Bioconductor library written in the R programming language. It generates various QC metrics and plots that relate duplication rate with gene expression levels in order to identify experiments with high technical duplication. A good sample with little technical duplication will only show high numbers of duplicates for highly expressed genes. Samples with technical duplication will have high duplication for all genes, irrespective of transcription level.

Credit: dupRadar documentation

Preseq

The Preseq package is aimed at predicting and estimating the complexity of a genomic sequencing library, equivalent to predicting and estimating the number of redundant reads from a given sequencing depth and how many will be expected from additional sequencing using an initial sequencing experiment. The estimates can then be used to examine the utility of further sequencing, optimize the sequencing depth, or to screen multiple libraries to avoid low complexity samples. A shallow curve indicates that the library has reached complexity saturation and further sequencing would likely not add further unique reads. The dashed line shows a perfectly complex library where total reads = unique reads. Note that these are predictive numbers only, not absolute. The MultiQC plot can sometimes give extreme sequencing depth on the X axis - click and drag from the left side of the plot to zoom in on more realistic numbers.

featureCounts

featureCounts from the Subread package is a quantification tool used to summarise the mapped read distribution over genomic features such as genes, exons, promotors, gene bodies, genomic bins and chromosomal locations. We can also use featureCounts to count overlaps with different classes of genomic features. This provides an additional QC to check which features are most abundant in the sample, and to highlight potential problems such as rRNA contamination.

DESeq2

DESeq2 is one of the most commonly used software packages to perform differential expression analysis for RNA-seq datasets.

This pipeline uses a standardised DESeq2 analysis script to get an idea of the reproducibility across samples within the experiment. Please note that this will not suit every experimental design, and if there are other problems with the experiment then it may not work as well as expected.

The script included in the pipeline uses DESeq2 to normalise read counts across all of the provided samples in order to create a PCA plot and a clustered heatmap showing pairwise Euclidean distances between the samples in the experiment. These help to show the similarity between groups of samples and can reveal batch effects and other potential issues with the experiment.

By default, the pipeline uses the vst transformation which is more suited to larger experiments. You can set the parameter --deseq2_vst false if you wish to use the DESeq2 native rlog option. See DESeq2 docs for a more detailed explanation.

Both types of transformation are performed blind, i.e. using across-all-samples variability, without using any prior information on experimental groups (equivalent to using an intercept-only design), as recommended by the DESeq2 docs.

The PCA plots are generated based alternately on the top five hundred most variable genes, or all genes. The former is the conventional approach that is more likely to pick up strong effects (ie the biological signal) and the latter, when different, is picking up a weaker but consistent effect that is synchronised across many transcripts. We project both of these onto the first two PCs (shown in the top row of the figure below), which is the best two dimensional representation of the variation between samples.

We also explore higher components in terms of experimental factors inferred from sample names. If your sample naming convention follows a strict policy of using underscores to delimit values of experimental factors (for example WT_UNTREATED_REP1) and all names have the same number of underscores (so excluding WT_TREATED_10ml_REP1 from being compatible with the previous label), then any of these factors that are informative (ie label some but not all samples the same) then we individually plot upto the first five PCs, per experimental level, for each of the experimental factors.

The plot on the left hand side shows the standard PC plot - notice the variable number of underscores, meaning that the central plot would not be produced: here we have changed the underscore that is hyphenating the treatment to a ’-’ character. This allows the central plot to be generated, and we can see that replicate (the 2nd part of the sample name) seems to be affecting the 3rd principal component, but the treatment factor is affecting the more important first two components. The right-most plot shows all pairwise euclidean distances between the samples.

Kraken2/Bracken

Kraken2 is a taxonomic classification tool that uses k-mer matches paired with a lowest common ancestory (LCA) algorithm to classify species reads. Bracken is a statistical method to generate abundance estimates based off of the Kraken2 output. These algorithms are run on unaligned sequences to detect potential contamination of samples. MultiQC reports the top 5 taxon members detected at the level of classification used for Bracken, with toggles available for higher taxonomic levels. If Bracken is skipped, MultiQC will report the top 5 species detected by Kraken2.

MultiQC

MultiQC is a visualization tool that generates a single HTML report summarising all samples in your project. Most of the pipeline QC results are visualised in the report and further statistics are available in the report data directory.

Results generated by MultiQC collate pipeline QC from supported tools i.e. FastQC, Cutadapt, SortMeRNA, STAR, RSEM, HISAT2, Salmon, SAMtools, Picard, RSeQC, Qualimap, Preseq and featureCounts. Additionally, various custom content has been added to the report to assess the output of dupRadar, DESeq2 and featureCounts biotypes, and to highlight samples failing a mimimum mapping threshold or those that failed to match the strand-specificity provided in the input samplesheet. The pipeline has special steps which also allow the software versions to be reported in the MultiQC output for future traceability. For more information about how to use MultiQC reports, see http://multiqc.info.

Pseudoalignment and quantification

Pseudoalignment

The principal output files are the same between Salmon and Kallisto:

An additional subset of files are distinct to each tool, for Salmon:

… and Kallisto:

As described in the STAR and Salmon section, you can choose to pseudoalign and quantify your data with Salmon or Kallisto by providing the --pseudo_aligner parameter. By default, Salmon is run in addition to the standard alignment workflow defined by --aligner, mainly because it allows you to obtain QC metrics with respect to the genomic alignments. However, you can provide the --skip_alignment parameter if you would like to run Salmon or Kallisto in isolation. If Salmon or Kallisto are run in isolation, the outputs mentioned above will be found in a folder named salmon or kallisto. If Salmon is run alongside STAR, the folder will be named star_salmon.

Transcripts with large inferential uncertainty won’t be assigned the exact number of reads reproducibly, every time Salmon is run. Read more about this on the nf-core/rnaseq and salmon Github repos.

The tximport package is used in this pipeline to summarise the results generated by Salmon or Kallisto into matrices for use with downstream differential analysis packages. We use tximport with different options to summarize count and TPM quantifications at the gene- and transcript-level. Please see #499 for discussion and links regarding which counts are suitable for different types of analysis.

According to the txtimport documentation you can do one of the following:

• Use the original (bias-uncorrected) counts with an offset: import all the salmon quant.sf files with tximport and then use DESeq2 with dds <- DESeqDataSetFromTximport(txi, sampleTable, ~condition) to automatically construct an object with gene-level offsets in-built, which can be used by DESeq2 to automatically account for effective gene length effects across conditions. It’s also possible to combine the .merged.gene_counts.tsv and .gene_lengths.tsv matrices output by this workflow to make your own object the same way - which is what the nf-core differentialabundance workflow does. See the DESeq2 module for the implementation details.
• Use bias-corrected counts without an offset: load and use salmon.merged.gene_counts_length_scaled.tsv or salmon.merged.gene_counts_scaled.tsv directly as you would with a regular gene-level counts matrix. These files were created using the tximport argument countsFromAbundance="scaledTPM" or "lengthScaledTPM" to scale counts to library size or to library size and average transcript length respectively.
• Use the original (bias-uncorrected) counts without an offset: load and use the txi$counts matrix (or salmon.merged.gene_counts.tsv) with DESeq2. This is generally not recommended, since it does not correct for potential differential isoform usage (the offset). However, if you have 3’ tagged RNA-seq data, then this is the most suitable method, because the counts do not exhibit any length bias.

NB: The default Salmon parameters and a k-mer size of 31 are used to create the index. As documented here and discussed here, a k-mer size off 31 works well with reads that are 75bp or longer.

Workflow reporting and genomes

Reference genome files

A number of genome-specific files are generated by the pipeline because they are required for the downstream processing of the results. If the --save_reference parameter is provided then these will be saved in the genome/ directory. It is recommended to use the --save_reference parameter if you are using the pipeline to build new indices so that you can save them somewhere locally. The index building step can be quite a time-consuming process and it permits their reuse for future runs of the pipeline to save disk space.

Pipeline information

Nextflow provides excellent functionality for generating various reports relevant to the running and execution of the pipeline. This will allow you to troubleshoot errors with the running of the pipeline, and also provide you with other information such as launch commands, run times and resource usage.