The PCAWG reference bundle
The studio's comparisons against PCAWG read one versioned bundle, built from its sources by a script that produces the same bytes every time (sigscape-reference/build.py). Version pcawg-v1:
| Table | Rows | What | |---|---|---| | tumors | 2,703 | aliquot, donor, specimen, project, tumor type (27), SNV counts per assay, DBS count | | sbs96_wgs | 2,703 | SBS-96 counts of the PCAWG consensus calls, PASS, tallied against hg19 | | sbs96_seqcap, sbs96_tcga_mc3, sbs96_msk | 2,703 each | the same tumors projected onto two exomes and a panel: their mutations inside each territory | | dbs78_wgs | 2,703 | DBS-78 counts (adjacent SNVs of one tumor are one doublet) | | exposures_musical | 2,703 | MuSiCal's PCAWG exposures (COSMIC v3.2), joined by specimen | | hr_status | 208 | HR status of PCAWG breast donors (58 deficient, 150 proficient) | | opportunity_hg19 | 32 | trinucleotide counts per territory |
Projection is the territory itself: a tumor's mutations inside an assay's targets, the subsetting SigMA2's simulate step applies before it resamples. Median counts per tumor: 5,257 SNVs genome-wide, 141 in the MC3 exome, 95 in the SeqCap exome, 5 in MSK-IMPACT's targets.
Sources, pinned
Every input is named with its sha256 in inputs.json and the build stops if one differs: the 2,703 consensus VCFs (as one hash over their hashes), MuSiCal's exposure table, the hg19 FASTA's index, three territory BED files, the HR-status list, and the PCAWG identifier table (aliquot to specimen to donor to project) from the IDConverter package at a fixed commit. --check rebuilds into a temporary folder and compares every file: identical.
What the browser gets
The site carries a slice (/reference/pcawg-v1/): the SBS-96 of every tumor with its project and tumor type (375 KB gzipped), the opportunity table, and the merged intervals of the exome and the panel for assay detection (1 MB gzipped). The vendor's SeqCap targets reach the browser only as trinucleotide counts. Everything is derived counts; no mutation leaves the build.
Not in this version
MuTopia per-tumor contributions and topotype mixes, Tau's genome-doubling calls and timing, purity and ploidy, and MuTopia's genome-wide component tracks are not on the machine that built pcawg-v1; they join in a later version through the same script.