SV inspector view
The SV inspector is a combined variant table and whole-genome circular view for triaging structural variant calls. For an end-to-end walkthrough on real cancer sequencing data, see the C-GIAB tutorial.
Launch it from the Add menu in the main menu bar; an import form then asks for your SV data.
The following formats are supported:
- VCF or VCF.gz (plain text or (b)gzipped)
- BED, BEDPE
- STAR-fusion result file
A STAR-Fusion table is a .tsv, so the import form reads its File Type off the
filename (star-fusion, fusion_predictions) or the #FusionName header line,
and the menu sets it by hand for a file named some other way;
Gene fusion calls and the DNA behind them walks a caller's output through the inspector
and out to the reads behind each call.
Record types and compatible callers
The SV inspector is best for long-range SV records: VCF entries with
SVTYPE=BND (breakends; the VCF-spec way to encode translocations) or the
caller-specific SVTYPE=TRA used by some tools.
Single-locus deletions and duplications load into the table like anything else, but the circular overview cannot draw them: a chord runs between a record's two ends, and at whole-genome scale a deletion's two ends are the same point. The legend below the circle says so — it counts every class in the rows on screen, including the ones with no chord to draw, so a mostly-local callset reads as such.
Compatible variant callers include:
- Short-read - Manta, Delly, Lumpy
- Long-read - pbsv, Sniffles
Example workflow
Load this VCF of translocation events called from the SKBR3 breast cancer cell line (published data). Paste the URL into the import form and select hg19:
SV inspector results
The loaded file appears as a searchable table, one variant per row, alongside a whole-genome circular overview.
Table filters are reflected in the circular view.
Rearrangement events
A complex rearrangement is several junctions, and some callers say which ones
belong together in VCF 4.4's EVENT key; the C-GIAB benchmark and DRAGEN both
write it. When the file has it, a Filter by event dropdown appears beside
Filter by SV type, listing each event with its record count and the
chromosomes it touches. Choosing one narrows the table and the circle to that
event's records, and with show only regions with data on, the circle redraws
over those chromosomes alone.
With no event chosen, selecting a row or chord of an event dims every chord outside that event, so its other junctions stand out against the whole callset.
The dropdown lists a value only when its records describe more than one
junction. A breakend callset writes each junction twice, once from either end,
and GRIDSS gives that pair an EVENT of its own, so a list of every value would
name every breakpoint in the file.
JBrowse reads the grouping and does not compute one, and it reads the standard
key only. A caller that files the same grouping under its own key needs the key
renamed before import. Severus
writes CLUSTERID, and MATE_ID where the specification has MATEID:
zcat severus_somatic.vcf.gz |
sed -e 's/^##INFO=<ID=CLUSTERID,/##INFO=<ID=EVENT,/' \
-e 's/^##INFO=<ID=MATE_ID,/##INFO=<ID=MATEID,/' \
-e 's/;CLUSTERID=/;EVENT=/' -e 's/;MATE_ID=/;MATEID=/' |
bgzip > severus_somatic.event.vcf.gzLaunching breakpoint split view
Click a feature in the circular view, or the triangle dropdown on any table row, to open the breakpoint split view for that variant: two stacked linear genome views, each centered on one breakpoint, both showing the callset. Add alignment tracks from the split view's track selector, and the arcs and splines connecting supporting reads appear automatically. The next row or chord opens in the same split view and keeps those tracks, including when its chain has more panels.
For a record filed under an event, the dialog offers Open every locus of that event: one panel per locus the event's junctions touch, in genome order, with breakends closer together than the window size sharing a panel.
To skip that step, name the evidence in the session or link: the
drilldownTracks launch key lists the trackIds
every view a row or chord opens starts with, such as the tumor and normal
alignments, a coverage track, or an assembly's synteny track.