Pangenome (HPRC) part 4, the graph as the picture
A pangenome graph holds shapes a linear track cannot draw: a repeat array is a knot of loops, a haplotype that skips sequence is an edge around it, and a region where dozens of haplotypes differ is a tangle. We draw the LPA kringle IV-2 array from the Human Pangenome Reference Consortium's release 2 graph as a graph, with each bubble named where it sits in the drawing, open the array to see its copies, and then lift one haplotype's walk out of the tangle to read how much sequence it carries through the array against GRCh38. Part 1 opened this graph beside the linear view and part 3 cut the eight haplotypes this page walks.
The graph view is a beta plugin, and this tutorial covers experimental ideas. Where a step below says the view works a particular way today, that is a current limit rather than a settled design. We welcome your feedback.
Prerequisites
- the GraphGenomeView plugin, loaded the way part 1 loads it
- htslib (
tabix), to query the hosted bubble index from the command line as the last section does
Where the data comes from
HPRC release 2, through the same rGFA projections part 1 uses and the eight-haplotype cut part 3 makes.
- the rGFA segment and link index of the SV-resolution graph, on GRCh38: https://jbrowse.org/demos/hprc/hprc-v2.1-mc-grch38.segs.bed.gz and https://jbrowse.org/demos/hprc/hprc-v2.1-mc-grch38.links.bed.gz
- the bubble index, one row per bubble from
gfatools bubble: https://jbrowse.org/demos/hprc/hprc-v2.1-mc-grch38.bubbles.bed.gz - the KIV-2 bubble cut from the base-level graph for GRCh38 and eight haplotypes, with their walks: https://jbrowse.org/demos/hprc/hprc-v2.1-mc-grch38.kiv2.eight-haplotypes.gfa
The window as a graph
Open the LPA window from part 1 and pick Force-directed layout from the Layout dropdown. The drawing is seeded along the reference, so GRCh38 runs left to right under the linear view in the same colours, and every bubble the index holds is drawn as a halo along its own nodes with a label saying what kind of variation it is.
The session's gene track is drawn onto the graph as well: LPA's name is pinned under the backbone at its midpoint, and its exons are the dark stretches along the reference nodes that carry them, so the array's backbone copies show the kringle exons they hold. An allele has no reference coordinates, so the loops carry none.
The halos are the bubble index drawn onto the graph. A deletion is a halo around the short arm of a bubble whose other arm is the backbone; an insertion is a halo around a loop hanging off one point; the array is the whole knot. Their labels carry the same numbers the bubbles lane above prints, so a bubble can be found in either picture from the other.
Open the array
Click the array's label. The view cuts the bubble's segments out of the graph on screen and lays them out on their own, with a button back to the window. No index row describes the inside of a bubble, so the popped graph derives its own bubbles from the way its nodes layer along the reference, and a bubble inside a bubble gets a label of its own to click in turn.
Every loop here is a different number of copies. The drawing says that much and nothing more: the rGFA records which segments exist and how they link, and no haplotype walks through it, so nothing in this file can say who takes which loop.
The same bubble with its walks
Part 3 cut this bubble from the base-level graph for GRCh38 and eight haplotypes, keeping each haplotype's walk. Open that file with Add → Graph genome view, or through the live link on the figure, and pick the force-directed layout again.
Two things change because the file carries walks. A node's thickness is the number of haplotypes through it, Bandage's depth drawn as width, so the backbone every haplotype shares is the thick line and the copies one haplotype alone carries are the thin loops. And the array's halo no longer counts routes through a DAG: its label is the shortest and longest walk between the bubble's two ends, which are real haplotype lengths. Each route over a kilobase also carries a chip at the far point of its loop naming the haplotypes that take it; with nine walks through one array those chips stack over the loops, so the figure below has Mark bubbles off.
Lift one haplotype out
The Walk dropdown in the toolbar names each haplotype in the file. Pick one and its route keeps its ink while every other node and link fades; a readout beside the legend gives the walk's length through the window and how it compares to the reference walk.
Lifting GRCh38 is the check. The reference takes no loop, so nothing but the backbone stays lit, and its readout compares the walk to itself. Lifting each of the other seven in turn reads each haplotype's copy count as an excess over the reference, one at a time, off the same drawing.
Check it against the index
The array's halo label in the first figure comes from one row of the hosted bubble index. Query it:
tabix https://jbrowse.org/demos/hprc/hprc-v2.1-mc-grch38.bubbles.bed.gz \
'GRCh38#0#chr6:160616002-160646753'The row's route count and its shortest and longest allele are what the label printed, and its segment list is what the pop cut out. The walk lengths in the last figure are not in this file at all: they come from the W lines of the eight-haplotype cut, which is why the array reads as one bubble of nine routes there and as a count of paths through the rGFA here.
See also
- Pangenome (HPRC) part 1, reading the graph
- Pangenome (HPRC) part 3, every haplotype in its own coordinates
- Pangenome (mouse), a bubble inside a bubble
References
- Liao WW, Asri M, Ebler J, et al. A draft human pangenome reference. Nature. 2023;617(7960):312-324. https://doi.org/10.1038/s41586-023-05896-x
- Wick RR, Schultz MB, Zobel J, Holt KE. Bandage: interactive visualization of de novo genome assemblies. Bioinformatics. 2015;31(20):3350-3352. https://doi.org/10.1093/bioinformatics/btv383
- HPRC release 2, the release this page opens.
Feedback on this tutorial is welcome: contact us.