# oncoanalyser, September 2026: T0 re-run and T2 organoid

nf-core/oncoanalyser **2.3.0** (Hartwig WiGiTS: REDUX, AMBER, COBALT, SAGE,
ESVEE, PAVE, PURPLE, LINX, LILAC, ISOFOX, NEO, CUPPA, ORANGE, …), WGTS mode,
genome `GRCh38_hmf`. Run on Google Cloud Batch on 2026-09-26: 248/248 tasks
succeeded in 12 h 49 min. One patient: IPISRC044 (UCLA sample ID SARC0277),
male.

This folder complements the April 2026 run in the parent directory
(`kamil/oncoanalyser/<group>/`), which has the T1 tumour, T1 organoid and T2
tumour groups.

## Groups

| folder | normal DNA | tumour DNA | tumour RNA |
|---|---|---|---|
| `IPISRC044_T0_personalis/` | Personalis blood WGS | Personalis tumour WGS, 2022-12-16 resection | BostonGene `BG003082`, same resection |
| `IPISRC044_T2_organoid_ucla/` | UCLA 2024 blood WGS | UCLA T2 organoid WGS (2025-02) | UCLA T2 organoid RNA-seq (2025-02) |

**Why T0 was re-run.** The April T0 run used Personalis RNA
`PSNLDx240603312`, which is a different, female donor. That April folder's
RNA-derived outputs are invalid; see its `SUPERSEDED.md`. This re-run uses the
same DNA plus the correct RNA.

**Reused alignments.** The normal and T0 tumour DNA were not re-aligned. The
April REDUX BAMs were given as input (`bam_redux`), and the reproduction is
exact: T0's PURPLE fit and its PASS somatic calls are identical to April's.
Only the T2 organoid WGS and both RNA libraries were aligned in this run.
`alignments/` therefore holds only the newly aligned BAMs; the reused ones are
in the April folders.

## QC summary

| | T0 (Personalis DNA + BostonGene RNA) | T2 organoid |
|---|---|---|
| PURPLE purity / ploidy | 0.65 / 3.25 | **0.18** / 3.15 |
| PURPLE QC | WARN_HIGH_COPY_NUMBER_NOISE (same as April) | WARN_LOW_PURITY |
| PASS somatic small variants / SV breakends | 3,704 / 196 | 1,219 / 68 |
| tumour WGS duplicates (REDUX) | — (April BAM) | **81%**: ~25x unique of ~150x raw |
| RNA: Isofox QC | PASS (73.3M fragments, 151 bp) | **FAIL_LOW_COVERAGE**: 97% duplicates, ~1.55M unique fragments |
| identity (somalier) | RNA matches IPISRC044 (rel 0.90–1.00, IBS0 ≤ 6) | WGS matches IPISRC044 (IBS0 0 vs the patient's WGS); RNA mostly does, but ~25% of it is another person's (see caveats) |

## Caveats for the T2 organoid

- **Low purity, as its parent tumour.** PURPLE puts it at 18% tumour, against
  16% for the T2 tumour itself (April run), and 34% vs 36% for the T1 organoid
  and tumour. Treat it as a mixed sample, not a pure tumour culture.
- **The organoid RNA is partly another person (added 2026-09-27).** A direct
  test of the reads at ~17,000 common SNPs finds 8.4% of them carrying alleles
  the patient doesn't have at sites where the patient is homozygous (patient
  RNA: 0.7%; a different person: ~32%), so roughly a quarter of this RNA comes
  from someone else. The WGS is not affected (PURPLE contamination 0). Treat
  every RNA-derived output of this group as unreliable: `isofox/`,
  `sage_append/`, the RNA parts of `neo/`, `cuppa/`, `orange/`.
- **Low-complexity libraries.** The WGS is 81% duplicates (~25x effective
  depth) and the RNA 97%. Isofox fails QC on the RNA, so its expression, its
  RNA support in `sage_append/` and `neo/`, and the RNA features in `cuppa/`
  are not reliable.
- **21 somatic calls are mouse artifacts.** The organoid WGS has ~1.3% mouse
  reads (culture material). Of its 256 PASS SNVs not seen in the T2 tumour, 21
  are supported only by reads that align better to mouse (mm39) than to human
  (`qc/organoid_mouse_artifact_calls.tsv`). Several come in pairs a few bases
  apart in conserved genes, the typical pattern. The 871 calls shared with the
  T2 tumour are clean (0 of 1,326 ALT reads mouse). Drop the 21 before using
  organoid-only variants.

## Layout

```
README.md                           this file
IPISRC044_T0_personalis/<tool>/     standard oncoanalyser output per tool
IPISRC044_T2_organoid_ucla/<tool>/
pipeline_info/                      Nextflow report, timeline, trace, params, software versions
qc/somalier_identity.tsv            somalier pairs: each new BAM vs 60+ IPISRC044 samples and the swapped donor
qc/organoid_mouse_check.tsv         per-variant mouse test (organoid-only low VAF, ≥10% VAF, shared control)
qc/organoid_mouse_artifact_calls.tsv  the 21 mouse-driven organoid calls to exclude
```

Each group's `orange/*.orange.pdf` is Hartwig's one-page summary report.

## Reproducing

The osteosarc project repository, `analysis/oncoanalyser/`:
`samplesheet.2026-09.csv`, `params.2026-09.yml`, `nextflow.config`, `run.sh`.
QC scripts: `check_organoid_mouse.py` there and
`analysis/somalier/run_somalier_2026-09.sh`.