Intron (splice-junction) track

GW can render a lightweight splice-junction track directly from the alignments already in view. For each BAM it extracts every N operation in the CIGAR of every visible read, clusters the donor and acceptor positions on each strand, and draws one bar per canonical intron — coloured and scaled by how many reads support it.

This makes it a quick way to get a splice-graph style overview of an RNA-seq BAM without pre-computing a BED, and it updates as you pan and zoom.

Showing the track

From the Tools menu, tick Toggle intron track under Introns.

From the command prompt:

:introns          # toggle for bam 0
:introns 1        # toggle for bam index 1

From the command line on startup:

gw hg38 -b rna.bam --introns

With multiple BAMs, --introns toggles the track on for every BAM that is loaded.

How junctions are built

  1. Harvest — for every read in view the CIGAR is walked for BAM_CREF_SKIP (N) ops. Each gap is recorded as a raw (donor, acceptor, strand) triple.
  2. Strand — the XS tag is preferred when present (Tophat/STAR output). When absent, the alignment’s reverse-strand flag is used so unstranded libraries still split into two separable buckets rather than mixing.
  3. Cluster — donor and acceptor positions are clustered independently on each strand with a 1D DBSCAN-style linear scan. Positions within splice_cluster_eps bp of each other collapse into one canonical coordinate (the mean of the cluster).
  4. Aggregate — every raw gap contributes one read of support to the (donor-cluster, acceptor-cluster, strand) it falls into.
  5. Filter — junctions supported by fewer than min_junction_reads reads are dropped.

Visual encoding

Each surviving junction becomes one bar:

  • Thickness grows with support, from thin at 1 read to thick at 30+.
  • Colour uses the matplotlib Purples ramp, clamped to the range 1..30 — anything above 30 reads uses the same darkest colour. On the igv theme light = low support; on dark and slate themes the ramp is reversed so light = high support on a dark background.
  • Strand is shown by the bar’s arrow: right-pointing for +, left-pointing for -.
  • Label — when the bar is wide enough, the supporting read count is drawn at the midpoint.

Settings

Both options are exposed in the Tools dialog under Introns and can be set on startup from the command line.

Option Default Meaning
splice_cluster_eps 2 Maximum bp distance between donor (or acceptor) positions merged into one canonical site
min_junction_reads 2 Minimum reads supporting a junction before it is drawn

Command-line equivalents:

gw hg38 -b rna.bam --introns --splice-cluster-eps 5 --min-junction-reads 3

Zoomed-out behaviour

Intron extraction runs over the alignments that GW has loaded into memory. When the view is wider than the low-memory threshold, GW switches to a coverage-only path and does not load individual reads — in that mode the intron track is cleared rather than estimated, so what you see is only ever built from reads actually in view. Zoom back in to repopulate.

Notes

  • The junction start/end coordinates use the cluster centroid, so two reads whose N ops begin at slightly different positions (e.g. because of aligner soft-clipping at the exon boundary) will still collapse into one intron when splice_cluster_eps is large enough.
  • Clustering is done per-strand, so + and - junctions never merge even if their coordinates happen to coincide.
  • The track is a synthetic track type — it has no backing file, and turning it off discards the extracted data.
This site uses Just the Docs, a documentation theme for Jekyll.