Splice Variant Caller
The identification of alternatively spliced isoforms (using their constitutive splice variants) and their functional effects is of high importance in the study of genetic variation and diseases, including cancer and neurological disorders. The main types of alternative splicing events resulting in splice variants are:
Exon skipping
Intron retention
Mutually exclusive exons
Alternative 5' splice site
Alternative 3' splice site
Running Splice Variant Caller
When enabled with the --enable-rna-splice-variant=true
option added to an RNA Map/Align job, DRAGEN runs a Splice Variant caller by taking advantage of its fast and highly accurate splice-aware read mapper/aligner that aligns to the whole genome to identify novel alternative Splice Junction (SJ) candidates. These candidates can be filtered by additional information provided such as a "normals list" and a "target regions list", or whitelisted with a "knowns" list.
Next during the read sorting phase, evidence for these alternative splices candidates (alts) vs. reference splicing are accumulated. Finally, each of the candidates are scored based on the accumulated read evidence and the results are written to TSV and VCF files for downstream tertiary analysis.
Following is an example command line.
Splice Variant Optional Input Files
In addition to the required inputs listed in the above example (i.e. paired fastq reads, reference hashtable, and annotation), the following 3 optional input resource files can be provided to help provide better precision by reducing FP count.
Normals List
A list of Normal splice variants that will be filtered out of the final output (i.e. operating as a blacklist), as long as they are not in the "knowns" list, using the "--rna-splice-variant-normals" option.
The format of this file should be a tab separated file in the same format as the SJ.out.tab, except only the first 4 columns are used, i.e.
contig name
first base of the splice junction (1-based)
last base of the splice junction (1-based)
strand (0: undefined, 1: +, 2: -)
To create a Normals list file, a collection of DRAGEN RNA mapper output SJ.out.tab files for at least 30 samples can be used along with a simple script to process all the SJs in these files. The pseudo code block below describes the function of this script:
Knowns List
A list of known splice variants that are exempt from being filtered out of the final output (i.e. operating as a whitelist), using the "--rna-splice-variant-knowns" option. The format of the file should be a tab separated file in the same format as the SJ.out.tab with 9 columns present, except only the first 4 columns are evaluated, i.e.
contig name
first base of the splice junction (1-based)
last base of the splice junction (1-based)
strand (0: undefined, 1: +, 2: -)
By default, the caller will not consider any splice variant candidates that are found in the input annotation file since it is looking for denovo variants, unless it is included in the knowns list which directs it not to discard the specified candidate. Note that some newer gene annotation models have added alt transcripts that contain clinically relevant splice variants, which causes the DRAGEN to skip reporting them.
To ensure these are reported, the user may want to pass these in with a knowns file containing these common variants if they are found in the annotation that is used. An example is shown below using hg38 coordinates specifying the MET exon 14 skip, EGFRv3, and ARv7 alt splicing events, respectively.
Target Regions BED
A list of regions that called splice variants must fall within using the "--rna-splice-variant-regions" option. Any splice variant candidates will be excluded if they are not within these regions.
This file should be in BED file format with the following info, except that the regions are 1-based
chromosome id
start position (1-based)
end position (1-based)
region (i.e. gene) name
Splice Variant Output Files
The detected splice variants are output as two separate TSV files for the intragenic and intergenic candidates, and as a VCF for the intragenic candidates.
The following categories are used when accumulating read counts for each alt SJ candidate:
DedupUniqueSupportingReads - Non-duplicate marked reads that are unique and precisely align with the SJ
DupUniqueSupportingReads - Duplicate marked reads that are unique and precisely align with the SJ
DedupUniqueNonsupportingReads - Non-duplicate marked reads that are unique but don't support the splice variant
DupUniqueNonsupportingReads - Duplicate marked reads that are unique but don't support the splice variant
To be counted, a paired read alignment:
must be primary and properly paired
must contain a splice junction (i.e. an alignment gap in the CIGAR containing skip ops)
must have overhangs on either side of the skip that are at least 6 bases
considered to be "unique" only if NH=1 and the MAPQ > 35
Splice Variant TSV Files
These two output files are named:
.splice_variants.tsv which contains the intragenic alt splice junctions that result in transcript variants
.splice_variant_fusions.tsv which contains the intergenic alt splice junctions that result fusions across genes
Each detected splice junction contains the following columns:
gene_start - Gene name(s) at the start of the SJ. Multiple genes are separated by a semicolon
gene_end - Gene name(s) at the end of the SJ. Multiple genes are separated by a semicolon
chromosome - Chromosome containing the SJ
start - SJ's start position (1-based genomic coordinate)
end - SJ's start position (1-based genomic coordinate)
strand - Detected strand for the SJ (1: +, 2: -)
motif - intron motif
annotated - True if annotated, otherwise False
split_unique_reads_ref - DedupUniqueNonsupportingReads count that support reference
split_total_reads_ref - DupUniqueNonsupportingReads + DedupUniqueNonsupportingReads count that support reference
split_unique_reads_alt DedupUniqueSupportingReads count that support variant
split_total_reads_alt - DupUniqueSupportingReads + DedupUniqueSupportingReads count that support variant
max_spliced_alignment_overhang - maximum spliced alignment overhang from all supporting reads
score - The splice junction variant score (ranging from 0.0 to 1.0). Currently, this is just a linear function of the number of split_unique_reads divided by 10, i.e. equals MIN(1.0, split_unique_reads_alt/10)
Note:
In the intragenic output file containing transcript variant splice junctions, the gene_start and gene_end columns must match.
In the intergenic output file containing fusions from splice junctions, the gene_start and gene_end columns must be different.
Splice Variant VCF File
This file contains the detected intra-genic splice junction variants that are not filtered out, and are written into a zipped VCF file titled .splice_variants.vcf.gz, where each splice variant candidate is written as a one-line VCF record containing the fields below:
CHROM - Chromosome of the splice
POS - SJ start position (1-based) i.e. first base of intron
ID - "." (unused)
REF - Base from the reference genome FASTA at the SJ start position
ALT - "<DEL>"
QUAL - The junction score from 0.0 - 1.0
FILTER - Semicolon separated list of filters: LowQ and LowUniqueAlignment
INFO - See the possible "Info fields below"
FORMAT - AD:DP
SAMPLE - Counts for {DedupUniqueSupportingReads}:{DedupUniqueNonsupportingReads}
The following lines of the VCF header describe columns 5 to 10 (last 6 columns)
Note on Filter Thresholds
The passing thresholds for the LowQ and LowUniqueAlignments filters are fixed to the settings below.
Filter | Description | Value |
LowQ | Below splice variant score threshold | less than 1.0 |
LowUniqueAlignments | Below unique supporting read count threshold | less than 2 |
Output Merged with Fusion Caller
When the splice variant caller and gene fusion caller are both enabled, the passing and failed intergenic fusion SJ's will also be merged into the relevant fusion output TSV files.
The passing calls get added to the fusion caller's .fusion_candidates.final file. The tab separated fields are described below.
Field Names | Description |
FusionGene | Left and Right gene names (separated by '--') |
Score | Value between 0 and 1 |
LeftBreakpoint, RightBreakpoint | The location for left and right sides of the splice with three colon separated fields: chromosome:coordinate:strand(+/-) |
Gene1Location, Gene2Location | Splice Variant caller always outputs "SpliceVar" here instead of Exon/Intron location |
Gene1Sense, Gene2Sense | Always TRUE for by design |
Gene1Id, Gene2Id | Long form ID (i.e. for Gencode it is usually "ENSG.version") |
NumSplitReads | Taken from the dedupUniqueSupportingReads count (i.e. split_unique_reads_alt column value) |
NumSoftClippedReads, NumPairedReads | These values are not used by RSV caller and are set to '0' |
ReadNames | Not provided by this caller and set to 'N/A' |
The failing calls get added into the fusion caller's .fusion_candidates.filter_info output file. The output fields are the same as described above for the "final" output file, with the addition of the FILTER_INFO field in the first column. The value in this field will be "RSV_FILTER:" followed by the specific filters that are not passing, as described in the table below.
Filter Names | Description |
LOW_QUAL | Below the "low quality score" threshold |
LOW_UNIQUE_ALIGNS | Number of unique anchors either on left or right side are below the MIN_UNIQUE_ALIGNS=2 threshold |
LOW_EVIDENCE_OR_OVERHANG | Not meeting the SJ.out read count vs. splice length and overhang requirements |
READTHROUGH | Gene partner is the next downstream annotated gene |
Last updated