SLIDR and SLOPPR identify spliced leaders (SLs) from 5'-tails of RNA-Seq reads that are soft-clipped after read alignment to a reference genome or transcriptome. SLIDR (Spliced leader identification from RNA-Seq data) assembles these read tails into full-length SLs and functional SL RNA genes. SLOPPR (Spliced leader-informed operon prediction from RNA-Seq data) searches read tails for a set of known SLs, quantifies SL-containing reads against all genes in the genome and uses SL usage patterns across genes to predict operons. SLOPPR can incorporate known SL specialisation for resolving downstream operonic genes (e.g., SL1/SL2-type SLs in nematodes), infer such specialisation de novo, or handle scenarios without SL specialisation.
Full descriptions of the implementation are detailed in the BMC Bioinformatics article: https://doi.org/10.1186/s12859-021-04009-7
Table of contents
# clone repository from GitHub
cd path/to/apps
git clone https://github.com/wenzelm/slidr-sloppr.git
# make scripts executable
chmod u+x slidr-sloppr/*
# add installation directory to PATH
export PATH=$PATH:path/to/apps/slidr-slopprThe R packages data.table, glmpca, ggplot2, scales, ggdendro, MASS and reshape2 are all available on CRAN and can be installed using install.packages() within the R console:
install.packages("data.table")
install.packages("glmpca")
install.packages("ggplot2")
install.packages("scales")
install.packages("ggdendro")
install.packages("MASS")
install.packages("reshape2")For convenience, both slidr.sh and sloppr.sh contain a function at the beginning of the script that loads dependencies:
function load_dependencies {
# ensure that all dependencies are in PATH
echo -e "#\n$(timestamp) >>> Loading dependencies"
module load hisat2-2.1.0
module load samtools-1.9
export PATH="$PATH:~/apps/ViennaRNA-2.4.14/bin"
...
}This is particularly useful when using an HPC. Please edit/remove the content of this function according to your machine configuration.
The script example_data.sh downloads the C. elegans genome assembly GCF_000002985.6 from NCBI, two very small RNA-Seq libraries from ENA (accessions ERR2756729 and ERR2756730) and runs basic SLIDR and SLOPPR analyses. Run this script to test the installation and to familiarise yourself with the workflow. Each analysis should complete within ten minutes using 8 threads and 32 GB of memory.
The script generates all input files and commands, supplying the genome assembly (-g), the genome annotations (-a), the output directory (-o) and a configuration file for multiple libraries (-m). The SLOPPR run also supplies a FASTA file containing the SL sequences (-s) and a text file specifying the SL2-type SLs (-S):
slidr.sh -g toy_data/GCF_000002985.6_WBcel235_genomic.fna -a toy_data/GCF_000002985.6_WBcel235_genomic.gff -o slidr_toy_data -m libraries_config.txt --agc
sloppr.sh -g toy_data/GCF_000002985.6_WBcel235_genomic.fna -a toy_data/GCF_000002985.6_WBcel235_genomic.gff -o sloppr_toy_data -m libraries_config.txt -s SL.fasta -S SL2.txtSL.fasta:
>Cel-SL1
GGTTTAATTACCCAAGTTTGAG
>Cel-SL2
GGTTTTAACCCAGTTACTCAAGSL2.txt:
>Cel-SL2The on-screen SLIDR results detail the expected SL1 and SL2 sequences, the numbers of reads assembled, the numbers of SL RNA genes identified and the numbers of trans-splice acceptor sites:
Consensus Reads SL_RNA_Genes SLTS_Sites
GGTTTAATTACCCAAGTTTGAG 6393 10 1234
GGTTTTAACCCAGTTTAACCAAG 146 1 64The on-screen SLOPPR results detail expectedly low SL-trans-splicing rates (5.14 %) and 84 predicted operons using SL2 as a polycistron resolver:
Numbers of genes receiving SLs:
not expressed 6399 32.28 %
no SL 12407 62.59 %
SL1 907 4.58 %
SL2 96 0.48 %
SL1+SL2 15 0.08 %
Predicted operonic genes, operons and operon sizes (n = genes in operon):
operonic operons n=2 n=3 n=4 n=5
SL2 180 84 75 7 1 1
Cluster1 1694 788 687 86 13 2
Cluster2 180 84 75 7 1 1Note that the SL-clustering algorithm may not correctly identify SL1/SL2-type SLs with only two ultra-low-coverage replicates. In the above run, the two clusters coincided with SL1/SL2 SLs (cluster2 = SL2). Due to some degree of stochasticity, your run may produce incorrect clusters.
# list all command-line options
slidr.sh -h
sloppr.sh -hSLIDR and SLOPPR share general options for data input and output:
-o <dir>
Path to output directory. If unspecified, the output directory is "./SLIDR[date+time]" or "./SLOPPR[date+time]"
-p <chr>
Name prefix for predicted SLs or operons (default: SL for SLIDR and OP for SLOPPR). It is recommended to follow a three-letter abbreviation of the organism, for example Cel for C. elegans.
-c <num>
CPU threads (default: 8. To use all available cores, enter $(nproc))
--tmp <dir>
Path to directory for temporary files. Default is your system's TMPDIR; specifying this option will change TMPDIR. For running SLIDR on large datasets, it is highly recommended to choose a large TMP directory to avoid potential bottlenecks associated with the default /tmp partition.
SLIDR and SLOPPR accept single-end or paired-end RNA-Seq reads in FASTQ(.gz) format.
Read alignments in BAM format are also accepted, for convenience of re-using alignments from previous pipeline runs. Please note that generic BAM files produced outside the pipelines without appropriate parameters for SLIDR or SLOPPR are NOT suitable as input! Please extract reads from such alignments in FASTQ format and then use these reads as input for the pipelines.
The following options are available to specify a single library:
-1 <file>
Path to R1 reads in FASTQ(.gz) or FASTA(.gz) format.
-2 <file>
Path to R2 reads in FASTQ(.gz) or FASTA(.gz) format.
-q
If specified, basic quality-trimming of 3' ends of reads and removal of Illumina adapters and poly-A/T tails is carried out using cutadapt -a AGATCGGAAGAGC -a 'A{100}' -g 'T{100}' -n 3 -q 20 -m 20 -l 150 -O 1. This is a useful convenience function but it is strongly recommended to carry out careful data filtering and trimming prior to running SLIDR or SLOPPR.
-b <file>
Path to pre-generated SLIDR/SLOPPR read alignments in BAM format. A BAI index file must be present in the same location.
-r <0|1|2|x>
Read strandedness generated during chemical library prep. This parameter is equivalent to the -s option in FeatureCounts:
0 = unstranded data (any read may originate from sense strand)
1 = forward stranded data (R1 reads originate from sense strand)
2 = reverse stranded data (R2 reads originate from sense strand)If strandedness is unknown, setting -r x (default) will infer it if genome annotations are supplied. If in doubt, unstranded analysis (-r 0) is always acceptable even for stranded data. Stranded data will produce less noisy results if the correct strandedness parameter is supplied.
Multiple libraries must not be specified with these options but instead via a configuration file supplied with the -m <file> option. This tab-delimited file must contain six columns:
-r option)-1 option)-2 option)-q option)-b option)Example file generated for the example dataset above:
ERR2756719 x toy_data/ERR2756719_1.fastq.gz toy_data/ERR2756719_2.fastq.gz trim
ERR2756720 x toy_data/ERR2756720_1.fastq.gz toy_data/ERR2756720_2.fastq.gz trimLeave columns empty if they are not required, for example:
Library1 <tab> x <tab> R1.trimmed.fq.gz <tab> R2.trimmed.fq.gz
Library2 <tab> 0 <tab> single_end.reads.fq.gz <tab> <tab> trim
Library3 <tab> 2 <tab> <tab> <tab> <tab> alignments.bamThis configuration file allows great flexibility in mixing single-end and paired-end libraries with different strandedness and existing read alignments.
Note: Avoid # characters in the file contents; they will be stripped!
Since version 1.2, SLIDR and SLOPPR can process libraries in parallel on HPC clusters. Default behaviour is sequential processing as in older versions.
--hpc
Switch on HPC mode. Automatic job submission and dependency control will be used to carry out the following workflow:
1) The main script generates the genome/transcriptome index as in sequential mode 2) For each library, a separate job is submitted, carrying out read alignment and SL tail extraction 3) A final dependent job is submitted, finishing the pipeline after all parallel library jobs have completed
The internal job submission control is designed for the SLURM job scheduler, but can be customised with the following options:
--hpcp <chr>
Command for submitting parallel jobs (one job per library). The default requests 4G of memory per specified thread (see -c). To modify, please specify the full submit command including all resources required including an option for returning the job ID in plain format:
--hpcp 'sbatch -c 4 --mem 8G --parsable' # SLURM with hard-coded resources
--hpcp 'qsub -v -cwd -pe smp 4 -l h_vmem=2G -terse' # Sun Grid Engine format
--hpcp 'msub -l nodes=1:ppn=4,pmem=8gb' # MOAB format--hpcd <chr>
Command for submitting the final dependent job. The default requests 8G of memory per specified thread (see -c). To modify, please specify the full submit command ending with the option for specifying a job dependency list for your HPC:
--hpcd 'sbatch -c 4 --mem 32G -d afterok:' # SLURM with hard-coded resources
--hpcd 'qsub -v -cwd -pe smp 4 -l h_vmem=8G -hold_jid ' # Sun Grid Engine format
--hpcd 'msub -l nodes=1:ppn=4,pmem=32gb -l depend=afterok:' # MOAB format--hpcs <chr>
Separator for job IDs in job dependency option. Default is ':' but some job schedulers (e.g., Sun Grid Engine) may require a comma instead (--hpcs ',')
If these options are not flexible enough to work with your HPC, you can still make use of parallel processing by doing the following:
1) Manually set up a task array job for aligning reads for all libraries (SLIDR:HISAT2 or BOWTIE2 commands; SLOPPR: hisat2 --no-softclip --no-discordant)
2) Use the BAM files instead of the raw reads as input for SLIDR/SLOPPR (-b option)
SLIDR requires either a genomic or transcriptomic reference assembly. A genomic reference is recommended and will produce vastly superior results. Genome annotations are recommended because they improve read mapping in HISAT2 and allow SLIDR to infer strandedness of the RNA-Seq data if unknown. SLIDR will also identify genes/transcripts in the annotations that overlap the predicted SL RNA genes and SLTS acceptor sites.
-g <file>
Path to genome assembly in FASTA(.gz) format.
-a <file>
Path to genome annotations in GFF(.gz) or GTF(.gz) format.
-t <file>
Path to transcriptome assembly in FASTA(.gz) format.
-T
Disable transcriptome trimming. Use this option if you want to use the supplied transcriptome as is.
NOTE: transcriptome trimming attempts to remove putative SL trans-spliced isoforms by clustering transcripts into genes and removing transcripts with 5' extension compared to other transcripts in each gene. Removing putative SLTS isoforms forces SL reads to align to non-trans-spliced isoforms of the gene and generate soft-clipped 5' read tails that are likely to contain the full 3' end of the SL sequence. These tails can then be aligned to the original untrimmed transcriptome to validate the SL sequence. The downside of this procedure is the aggressive loss of alternative non-trans-spliced isoforms, which may introduce alignment errors across exon-exon junctions.
-l <num>
Minimum length of soft-clipped tails to retain after alignment (default: 8). If very few reads pass filters it may be useful to decrease this value.
-x <num>
Scale factor for upper tail length limit (default: 1.0). This factor controls the amount of soft-clipping allowed during read alignment in HISAT2 or BOWTIE2. This is achieved via the minimum score functions:
hisat2 --score-min L,5,-0.4*x --sp 1,0 --mp 3,1
bowtie2 --score-min L,5,1-(0.4*x) --ma 1 --mp 3,1 --local--sp 1,0), the minimum score will be equivalent to the maximum number of soft-clipped bases. The above formula thus yields a minimum score of 5 - 0.4*x*readlength = -35 for a 100 bp read and -x 1.0, or a maximum of 35 soft-clipped bases. Assuming no mismatches, no gaps and no 3' soft-clipping, these settings can thus capture an SL tail of up to 35 bp. In practice, this length is reduced due to mismatches, gaps and 3' soft-clipping caused by poor-quality read ends or adapter contamination, but empirically works well for capturing nematode-sized SLs (c. 22 bp) even from 75 bp reads. --ma 1), a perfectly aligned 100 bp read would have a score of 100. Allowing an SL tail of up to 35 bp, the minimum score would have to be 100 - 35 = 65, which corresponds to the above formula 5 + (1-(0.4*x))*readlength = 65Modifying the scale factor -x can restrict or expand the upper tail length limit according to the formulae above and illustrated in this table (rows = read length; columns = x; values = limit):
x 0.25 0.5 0.75 1.0 1.25 1.5 1.75 2.0
---------------------------------------------------------------------
50bp 0 5 10 15 20 25 30 35
75bp 3 10 18 25 33 40 48 55
100bp 5 15 25 35 45 55 65 75
125bp 8 20 33 45 58 70 83 95
150bp 10 25 40 55 70 85 100 115--agc
Cluster read tails using abundance-based greedy clustering (AGC) instead of the default distance-based greedy clustering (DGC) (what's that?). Use this option if DGC yields poor coverage in the final SLs. See also the parameter guidelines for more information.
-e <num>
BLASTN E-value (default: 1). This parameter controls the stringency of the alignment of the read tail cluster centroids against the reference assembly. The default value of 1 is appropriate for centroid lengths as low as 10 bp and should rarely require modifying. Decreasing the value will require longer tail matches and thus substantially reduce the numbers of reads passing filteres. Increasing the value will allow shorter centroids to be retained, which we found necessary when analysing the short 16bp SL in Ciona intestinalis. Note that increasing this value will substantially increase the numbers of sequence alignments to process.
-D <chr>
Splice donor site pattern in regex notation (default: GT). Alternative nucleotides can be coded with character classes, for example, -D 'G[TC]' matches GT or GC, and -D 'A[AG][TC]' matches AAT, AAC, AGT or AGC. To switch off, specify empty character string (-D '').
-A <chr>
Splice acceptor site pattern in regex notation (default: AG). Alternative nucleotides can be coded with character classes, for example, -A 'A[CG]' matches GT or GC. To switch off, specify empty character string (-A '')
-S <chr>
Sm binding site motif and location in regex notation. This allows for searching additional motifs (Sm or otherwise) downstream of the splice donor site. The default (.{20,60}AT{4,6}G) matches the Sm binding sites ATTTTG, ATTTTTG or ATTTTTTG 20-60 bp downstream of the splice donor. Any custom regex patterns are supported, for example, -S '.{20,60}AT{4,6}G.{20,30}T{3,5}' to add a T-rich region 20-30 bp downstream of the Sm binding site. To disable, specify empty character string (-S '').
-R <chr>
Maximum SL RNA length excluding SL (default: 80). This length is measured starting from and including the splice donor site. The default of 80 bp is appropriate for nematode SL RNAs that are c. 100 bp long including a c. 22 bp SL.
-O <chr>
Maximum overlap between the 3' end of the SL and the trans-splice acceptor site (default: 10). If a transcriptome reference is used, it might be prudent to decrease this value or even set it to 0 if the transcripts do not contain extra non-coding 5' bases; this will reduce noise.
-L <chr>
Maximum base-pair span within stem-loop (default: 35). This parameter controls stem-loop prediction in RNAFold. The default requires the first and last base of a stem loop to be no more than 35 bp apart.
Log files and intermediate output files are written to the directories 1-library_[library name], 1-tails and 2-RNA-filters, representing each main pipeline stage.
Final results are written to the directory 3-results-[suffix] inside the specified output directory. The suffix of the directory name summarises the specified parameters to allow for convenient parameter sweeps within the same output directory, for example slidr_toy_data/3-results-x1.0-l8-AGC-e1-R80-DGT-S.{20,60}AT{4,6}G-L35-AAG-O10
Main output:
SL.summary.tsv: tab-delimited table summarising SL sequence, read coverage, numbers of SL RNA genes, numbers of SL trans-splice acceptor sites, numbers (median and range) of stem loops and spatial structure stability statistics from RNAFold (MFE frequency and ensemble diversity). The spatial structure statistics could be used to manually rank SL candidates by plausibility (higher MFE frequency is better; lower ensemble diversity is better).SL.sequences.fa: all SL sequences in FASTA formatSL.RNA_genes.fa: all SL RNA gene sequences in FASTA formatSL.RNA_genes.gff3: all SL RNA genes in GFF3 formatSL.acceptor_sites.gff3: all SL trans-splice acceptor sites in GFF3 formatraw.summary.tsv: same as SL.summary.tsv, but including singleton SLs (defined by only a single read and/or spliced to only a single acceptor site). raw.tails.tsv: summary of read depths of each unique read tail and numbers of read alignment locations. These are raw statistics prior to SL assembly and SL RNA filters; these might be useful for exploration if no good SLs can be found, particularly for runs with a reference transcriptome instead of a genome.Output per SL (in SLs subdirectory):
SLs/*.RNA_genes.fa: SL RNA gene sequences (FASTA) for each SLSLs/*.RNA_genes.gff3: SL RNA gene sequences (GFF3) for each SLSLs/*.acceptor_sites: SL trans-splice acceptor sites (GFF3) for each SLTo avoid writing potentially thousands of output files, SL-specific output files are written only for the first 100 SL candidates. If this is not enough, the limit can be increased manually in the script slidr_consensus.R (line 4: filemax <- 100). Data for specific SLs could also easily be extracted from the main output files using grep or awk.
If genome references annotations were supplied, an additional output file (*.reference.txt) is produced for each GFF3 output file, containing overlaps between each GFF3 entry and the genes/transcripts of the reference genome annotations (output of bedtools intersect). Note that SLTS acceptor sites are extended downstream by 100 bp to search for overlaps. This is a convenience function that does not replace careful manual curation of SL RNA genes and SLTS acceptor sites in a genome browser.
SLOPPR requires a genomic reference assembly and genome annotations. If annotations are not available, we recommend generating de novo annotations from the RNA-Seq data using BRAKER2 or STRINGTIE.
-g <file>
Path to genome assembly in FASTA(.gz) format.
-a <file>
Path to genome annotations in GFF(.gz) or GTF(.gz) format.
-s <file>
Path to SL sequences in FASTA format. Ensure that each SL has a unique name (FASTA header).
-n <num>
Minimum bp from 3' end of SL required to detect SL tail in read (default: 8). Increase this value if the SLs cannot be distinguished reliably at their 3'-most 8 bases. Note that longer tails will substantially reduce the number of SL reads recovered.
-e <num>
Maximum error rate for SL tail matching (default: 0.09). The default error rate requires no mismatches for tails up to 10 bp total length and allows for 1 mismatch for each 10 bp additional length (0-10 bp: 0; 11-21 bp: 1; 22-32 bp: 2; 33-40 bp: 3).
-f <chr>
GTF feature ID field used for counting reads (default: exon). If the genome annotations miss exon features, this option should be set to CDS.
-F <chr>
GTF meta-feature ID field used for summarising read counts (default: gene_id). If the genome annotations do not define genes, this option should be set to transcript_id.
-z [geo|sum|median]
Method for aggregating SL counts across libraries (default: geometric mean -z geo). The geometric mean is an appropriate statistic for summarising count data. We also implement sum and median, the latter of which may be more appropriate if giving weight to libraries with zero counts (see below).
-0
Keep libraries with zero counts when aggregating SL counts (default: remove zeros). When keeping zeros, we recommend aggregating with the median instead of the geometric mean. The geometric mean will drop all counts if at least one libraries has zero counts - such a stringent filter may be useful for removing lowly and inconsistently SL trans-spliced genes.
-S <file>
Optional file containing the names or FASTA headers of SLs (from those supplied by -s) that are specialised for resolving polycistrons (SL2-type SLs). Omit this option if SL specialisation is unknown or absent. Note that SLOPPR will additionally infer SL1/SL2-type subfunctionalisation irrespective of -S and do additional operon prediction runs using these inferred SL clusters. As a sanity check, the two inferred clusters should coincide with known SL1/SL2-type SLs. If SL2-type SLs are unknown, the two clusters represent the most likely candidates for SL1/SL2 types.
-d <num>
Minimum SL2:SL1 read ratio required to classify a gene as downstream operonic (default: infinity, i.e. no SL1 reads). SLOPPR will carry out three independent operon prediction runs:
-SMany organisms with SL2-type SL specialisation do allow some degree of SL1-type trans-splicing at downstream operonic genes, so it is worthwhile to relax this option (e.g., -d 2)and tune it based on the genome-wide distribution of SL2:SL1 ratios.
-u
Enforce the same SL2-type bias at upstream operonic genes as at downstream operonic genes. By default, SLOPPR requires the upstream gene of an operon to have SL1-type bias or not to be SL trans-spliced at all. If no such gene is available, SLOPPR designates the first downstream gene as an ad-hoc upstream gene; operons with an "adhoc" upstream gene are flagged as provisional in the output because they violate the assumptions. The -u option results in a much stricter set of operons where all genes have SL2-bias.
-i <num>
Maximum intercistronic distance in predicted operons (default: infinity; x = infer). By default, SLOPPR predicts operonic genes solely based on SL2-bias and ignores intercistronic distances. This may mean that some operonic genes have unrealistically large intercistronic distances. Such false predictions can be avoided by supplying a fixed cutoff (for example, -i 100) or using automatic inference of the most likely cutoff (-i x) given the distribution of intercistronic distances among the initial set of operonic genes. Filtering by intercistronic distance is particularly important in organisms where no SL specialisation exists; in these situations one must tease apart operonic from monocistronic SL-receiving genes by exploring intercistronic distance distributions among SL trans-spliced genes.
-x <path>
Reference operon annotations (GFF/GTF). If supplied, SLOPPR will examine predicted operons for overlap with these reference operons. Note that the GFF/GTF must contain only a single record per operon that spans all operonic genes. Do not include "gene" entries for individual operonic genes. This is a convenience option that is not designed to replace thorough synteny analysis.
Genome-aligned SL reads (BAM format) merged from all libraries are written to the subdirectory 1-merged_SL_BAM.
Read quantification results against gene annotations are written to the subdirectory 2-counts:
bg.featureCounts.genes.*: featureCounts results from initial end-to-end read alignments (background gene expression)un.featureCounts.genes.*: featureCounts results from unsuccessful candidate reads (background gene expression)SL.featureCounts.genes.*: featureCounts results from SL readsSL.featureCounts.exons.*: featureCounts results from SL reads, quantified against curated gene annotations, where genes were split at internal exons receiving SL reads*.raw.txt versions are raw featureCounts count tables and *.clean.txt versions are simplified featureCounts tables used for analysis.
Operon prediction results are written to the subdirectory 3-operons-[suffix]. The suffix of the directory name summarises the specified parameters to allow for convenient parameter sweeps within the same output directory, for example sloppr_toy_data/3-operons-zgeo-0remove-dinfinity-uno-iinfinity. All output files are written both for the SL.featureCounts.genes and the SL.featureCounts.exons featureCounts tables:
*.SL_clusters.txt: inferred SL subfunctionalisation cluster and PCA/LDA coordinates for each library*SL combination*.SL_dendrogram_groups.txt: hierarchical clustering memberships of each library*SL combination*.SL_genes.txt: library-specific and overall frequencies of genes receiving any combination of SL types (SL1, SL2, SL1+SL2) or clusters (Cluster1, Cluster2, Cluster1+Cluster2)*.SL_readratio.txt: quartiles of SL2:SL1, Cluster1:Cluster2 and Cluster2:Cluster1 read ratios across genes that receive both SL types.*.operons.summary.txt: numbers of operons, operonic genes and distribution of operon sizes, using either SL2, Cluster1, or Cluster2 as polycistron resolvers*.operons.intergenic_distances.txt: quartiles of intercistronic (=operonic) and intergenic (=non-operonic) distances, using either SL2, Cluster1, or Cluster2 as polycistron resolvers*.operons.[SL2|Cluster1|Cluster2].txt: gene names, locations, distances, background gene expression (meanTPM), SL1/SL2 read counts (CPM), SL2:SL1 read ratios, inferred operonic status and quality (pass or adhoc; see -u option) of all genes, using either SL2, Cluster1, or Cluster2 as polycistron resolvers*.operons.[SL2|Cluster1|Cluster2].gff3: predicted operons and operonic genes in GFF3 format, using either SL2, Cluster1, or Cluster2 as polycistron resolvers. Operons are flagged as quality:provisional if they contain adhoc upstream genes (see -u option)Most of these files are also available in graphical format as PDF. Log files and intermediate output files are written to the directories 1-library_[library name] and 2-counts.
Yes, genome annotations are not mandatory and SLIDR will function fine without them. However, genome annotations are useful for three reasons:
Note that all SLIDR results are solely based on read alignments and do not take gene annotations into account.
Yes, you can assemble transcripts (for example, using TRINITY) and supply these as a reference assembly to SLIDR:
slidr.sh -t Trinity.fastaNote that SL RNA filters (-D and -S) will only work if your assembly contains SL RNAs. Your mileage may vary!
If your organism is a nematode, run SLIDR with default parameters, which are optimised for nematode SLs. If your organism is a different eukaryote, modify or switch off the Sm binding motif filter (-S '') and allow longer read tails (-x 1.5) to capture a broad range of tail lengths:
slidr.sh -S '' -x 1.5Use this initial SLIDR output to build a more stringent set of filters for a second run, informed by sequence features of the most promising candidates (i.e, those with high read coverage and spliced to a large number of genes). If you need to relax SLIDR further, consider modifying (or even disabling) splice donor/acceptor site motifs (-D, -A) and/or increasing the BLASTN E-value (-e). An extremely relaxed configuration would be as follows (not recommended because of potentially huge data volumes to process):
slidr.sh -D '' -S '' -A '' -e 5 -x 1.5In our experience, SLIDR often detects novel SL variants and in some cases perhaps even novel SL classes. SLIDR is also useful for telling apart functional SL RNA genes from pseudogenes. We recommend SLIDR as a tool for initial data exploration to uncover untapped SL diversity even if the canonical SL sequences are already known.
The safest way is to run SLIDR multiple times, each time optimising parameters for each SL type. Alternatively, try and define parameters such that they capture all SL types at once. For example, in Hydra vulgaris where two major SL classes exist with different SL lengths and Sm binding motifs, the following parameters capture both SL types:
slidr.sh -S '.{10,35}[AG]ATTTT[CG][AG]' -x 1.4 -R 60SLIDR is not designed to find all possible SL RNA genes in a genome. SL RNA genes must be expressed, i.e. the SL encoded by the gene must be detected as a read tail and pass length filters. Similarly, the SL RNA gene must satisfy the splice donor, splice acceptor and Sm binding motif filters. This means that SLIDR will report all possible expressed SL RNA genes given the RNA-Seq libraries, but will not report unexpressed genes or genes not satisfying functional motif filters (since these may be pseudogenes). If a comprehensive annotation of putative SL RNA genes is required, SLFinder is a more appropriate tool.
Consider relaxing or disabling nucleotide motif filters to test whether this yields more genes for the focal SL, for example:
slidr.sh -D 'G[TC]' -S '' -A 'A[GC]'That depends on your dataset and is impossible to know in advance. To illustrate the difference between the two methods, consider the following tails after dereplication:
Tail Abundance
1: TTAGCTTAGCAGTAGGGGAGTTTGAG 1
2: GGTTTAATTACCCAAGTTTGAG 10
3: AATTACCCAAGTTTGAG 100
4: ACCCAAGTTTGAG 1000
5: AGTTTGAG 10000Note that tail 2 is the full-length C. elegans SL1 sequence and tails 3-5 are 5' truncated versions that clearly should be clustered with tail 2. However, the short and highly abundant tail 5 also happens to be identical to the 3' end of tail 1, which only appears once and is likely to be noise.
Using DGC, ties are broken by sequence length, so tail 5 will cluster with tail 1 instead of tail 2, yielding the following two clusters:
Cluster centroid Abundance
1: TTAGCTTAGCAGTAGGGGAGTTTGAG 10001
2: GGTTTAATTACCCAAGTTTGAG 1110Using AGC, ties are broken by abundance, so tail 5 will cluster correctly with tail 2 because it is more abundant:
Cluster centroid Abundance
1: TTAGCTTAGCAGTAGGGGAGTTTGAG 1
2: GGTTTAATTACCCAAGTTTGAG 11110That's a dramatic difference in read coverage! So is AGC always superior to DGC? No! Let's remove tail 1 and add another unspecific tail X that happens to be highly abundant:
Tail Abundance
2: GGTTTAATTACCCAAGTTTGAG 10
3: AATTACCCAAGTTTGAG 100
4: ACCCAAGTTTGAG 1000
X: TAGTTTGAG 5000
5: AGTTTGAG 10000Using AGC, tail 5 will now cluster with tail X instead of tail 2 because its abundance is higher (5000 vs. 1110):
Cluster centroid Abundance
2: GGTTTAATTACCCAAGTTTGAG 1110
X: TAGTTTGAG 15000Conversely, using DGC, tail 5 clusters correctly with tail 2 because it is longer:
Cluster centroid Abundance
2: GGTTTAATTACCCAAGTTTGAG 11110
X: TAGTTTGAG 5000It is obvious that the two clustering methods are bound to yield very different results in organisms with many SL variants that happen to be conserved at the 3' end. In those cases, short tails will match multiple SLs and ties must be broken arbitrarily (length or abundance). Most datasets we have analysed yield better results with the default DGC, but others performed poorly and improved dramatically with AGC. We therefore suggest using the default DGC and trying AGC if SL read coverage is suspiciously low.
Yes, but be aware of bottlenecks:
SLIDR typically does not need huge amounts of raw data to yield robust results, so consider starting with a small number of libraries.
SLOPPR finds operons by designating downstream operonic genes via SL2-type bias; upstream genes are then added to each tract of downstream operonic genes. Therefore, you must identify what SLs your organism uses to resolve downstream operonic genes. This may be a specialised set of SLs that are not usually added to any other genes, or it may be the same SLs that are also added to upstream operonic or monocistronic genes. SLOPPR can model virtually any scenario, but requires careful specification of several parameters (-S, -u, -d, -i)
This is the default SL1/SL2-type scenario encountered in C. elegans and many other nematodes. It is a good place to start even if your organism is not a nematode. Simply supply a text file containing the names of the SL2-type SLs with the -S option:
sloppr.sh -s sl_sequences.fasta -S sl2-type.txtSLOPPR will designate downstream operonic genes via SL2-type bias and disallow SL2-type bias for upstream operonic genes where possible. Relax the SL2:SL1 ratio (-d option) if you want to allow some SL1-type reads at downstream genes, and also use the -i option to set a maximum intercistronic distance (e.g., 500 bp) if appropriate:
sloppr.sh -s sl_sequences.fasta -S sl2-type.txt -d 2 -i 500Note that SLOPPR also carries out additional prediction runs using inferred SL types instead of those supplied with -S. Use the SL clustering results and the predicted operons is as a sanity check to verify that the assumed SL1/SL2-type SLs are plausible. If SLOPPR's inferred SL clusters do not coincide with the assumed SL1/SL2 types, consider that you may have mis-specified your SLs or your SL types may not be as well characterised as expected.
Omit the -S option to automatically designate all SLs as "SL2-type":
sloppr.sh -s sl_sequences.fastaSince SL2:SL1 ratios are undefined in this case, every gene receiving any SL will be designated as downstream operonic. Each operon will also receive an upstream gene that is not SL trans-spliced (if possible). By definition, no monocistronic SL-receiving genes are designated.
Use the inferred SL clustering as a sanity check to verify absence of SL subfunctionalisation.
As above, but use the -u option to enforce SL2-type bias at upstream operonic genes:
sloppr.sh -s sl_sequences.fasta -uOmit the -S option and carefully inspect the SL clustering results for plausible SL1/SL2-type subfunctionalisation:
sloppr.sh -s sl_sequences.fastaIn this situation, both operonic and monocistronic genes receive the same SLs. Instead of using SL2:SL1 read ratios, we must tease out operonic genes via short intercistronic distances from monocistronic genes with large intergenic distances.
First, omit the -S option and observe vastly inflated intercistronic distances (because every gene receiving any SL will be designated as downstream operonic):
sloppr.sh -s sl_sequences.fastaThen, run the automatic intercistronic distance filter, or specify a cutoff directly (e.g., 500 bp):
sloppr.sh -s sl_sequences.fasta -i x
sloppr.sh -s sl_sequences.fasta -i 500These filters will cause SL-receiving genes with large intergenic distances to be designated as "monocistronic".
Omit the -S option and ignore all SL clustering results:
sloppr.sh -s sl_sequences.fastaAs above, but use the -u option to enforce SLs at upstream operonic genes:
sloppr.sh -s sl_sequences.fasta -uAs above, but use intercistronic distance filtering to designate monocistronic genes:
sloppr.sh -s sl_sequences.fasta -i x
sloppr.sh -s sl_sequences.fasta -i 500Yes, but be aware of bottlenecks:
new versions SLIDR 1.2 and SLOPPR 1.2:
new version SLIDR 1.1.6: fixed error when reading empty columns with data.table
fixed ggplot2 bug in figure legends (SLOPPR). Results are unaffected.
new versions SLIDR 1.1.5 and SLOPPR 1.1.5: fixed errors when BAM files are specified as input files
new version SLOPPR 1.1.4: fixed an error and a silent bug triggered in some circumstances by SL names that are contained in other SL names
Please cite the BMC Bioinformatics article:
Marius A. Wenzel, Berndt Mueller, Jonathan Pettitt. SLIDR and SLOPPR: Flexible identification of spliced leader trans-splicing and prediction of eukaryotic operons from RNA-Seq data. Bioinformatics 22, 140 (2021), doi: https://doi.org/10.1186/s12859-021-04009-7