OncoTracer 2.2.0 RC1: GPU execution and read-backed linkage evidence

This preview adds GPU selection in browser setup and resumed analyses, verified optional Conda GPU environments, and a Docker image with compatible TensorFlow and PyTorch runtimes. GPU-capable ONT callers and local models share a device lock and retain VRAM headroom for the desktop. Alignment, CNA, GLIMPSE2, FFPErase and Sturgeon remain CPU tasks. See the GPU guide.

The new opt-in oncotracer haplotype-evidence command assesses linkage in completed BAM/VCF files. It reports Monte Carlo and exact conditional linkage p-values, dependent-test correction, and explicit insufficient-evidence results. It does not estimate germline or true-variant probabilities. Optional VarSome API queries retrieve sourced annotations; the local LLM can select only literal annotation fact IDs. See the method and commands.

Validation: 1,389 main tests (3 skipped), 128 GLIMPSE tests, copied executable and package checks, 10 hosted Conda solves, and local Docker/GPU checks passed. Real-region Clair3 and ClairS-TO CPU/GPU comparisons retained the same variant keys, genotypes and filters, with small numerical QUAL differences. The Docker image was pulled back and its identity verified. Details are in runtime-validation.json.

This is a prerelease. The historical frozen-CNA parity gate remains unresolved. Hosted Docker CI stopped because its runner had about 85 GiB free and the build requires 96 GiB; the local image checks passed, but overall hosted CI is not green. Stable release gates remain intact. PyPI and Docker latest have not been updated.

Install this exact preview:

python3 -m pip install --upgrade https://github.com/cfarkas/oncotracer/releases/download/v2.2.0-rc.1/oncotracer-2.2.0-py3-none-any.whl
oncotracer install --conda --gpu
oncotracer doctor --backend conda --gpu
oncotracer --interactive --backend conda

Docker on a host with NVIDIA Container Toolkit:

docker pull carlosfarkas/oncotracer:2.2.0-rc.1
oncotracer install --docker --image carlosfarkas/oncotracer:2.2.0-rc.1
oncotracer doctor --backend docker --gpu
oncotracer --interactive --backend docker

Select GPU usage in the browser. Native/Nextflow CPU usage remains available. Docker digest: sha256:a7b110689803dd1f8ea0336668dd79213828df04793574c9c44e0f429cad635e (linux/amd64). Source: 1fe0f11c9b68. CI results.

OncoTracer 2.1.1 RC1: verified parallel Nextflow startup

Parallel samples could read the Nextflow runtime JAR while another process was still copying it into a shared cold cache. The production launcher and unmodified JAR reproduced 5 failures among 26 simultaneous starts. This repair prepares and verifies the exact worker runtime once under a process lock, then atomically publishes an immutable generation before scheduling samples. Per-JVM temporary/native-library isolation is retained.

The release also prevents Python bytecode from contaminating verified standalone payloads, preserves the tool-inventory API, and supports verified reuse of unchanged 2.1.0 Conda environments.

Validation: 1,486 unit/audit tests passed, 111 repaired host starts, 16 Docker cold-cache starts, and 2 whitespace-path starts passed. The copied executable, wheel, Conda doctor and 17 Docker checks passed. The registry image was pulled back and verified against the tested image. Detailed counts and identities are in runtime-validation.json; download checksums are in SHA256SUMS.

This is a prerelease. The frozen-v1.1 CNA comparison is unresolved after intentional low-coverage rejection and segmentation changes. Its thresholds have not been relaxed, and the stable-release gate remains intact. Startup tests do not establish variant sensitivity, full patient-run completion, or the cause of every historical JVM SIGSEGV. PyPI has not been updated by this release.

Native installation of the verified wheel from this revision:

python3 -m pip install --upgrade "https://github.com/cfarkas/oncotracer/releases/download/v2.1.1-rc.1/oncotracer-2.1.1-py3-none-any.whl"
oncotracer --help
oncotracer install --conda
oncotracer doctor --backend conda
oncotracer --interactive --backend conda

Docker image:

docker pull carlosfarkas/oncotracer:2.1.1
docker run --rm carlosfarkas/oncotracer:2.1.1 doctor --backend host

Image digest: sha256:2cf033ab24939e427d53a0eb206561f1ad44057a2dee1244c10bd51a7bbc19fa (linux/amd64). Source: ba3985a0481f. Native CI.

OncoTracer v2.1.0

OncoTracer v2.1.0

Local browser setup and results dashboards, clearer matched HTML/PDF knowledge reports,
and safer setup/uninstall recovery. The copied executable runs the native workflow
without a Nextflow dependency.

• Choose ONT or Illumina, review detected FASTQs, label samples, and save or run from the local browser.
• Open organized result dashboards with final CNA tables, plots, reports, and retained diagnostics.
• Knowledge HTML/PDF reports and their shared sample indexes are under 04_cna_custom_plots/llm_reports/.
• Existing v2.0.0 releases and versioned tool installations are preserved.

Validated on exact commit: 5bb0646b1d1b1b923cdb9d47fbbb71c1ad6f4cb4

• Native v2 CI passed on the exact release commit.
• QuickStart 1 passed semantic v1.1/native-v2 parity for Illumina and ONT.
• QuickStart 2 passed semantic v1.1/native-v2 parity for all three HCC1143 libraries.
• The frozen comparator uses OncoTracer v1.1, SAMURAI v1.4.0, pinned Nextflow, and immutable container digests.
• Container: ghcr.io/cfarkas/oncotracer@sha256:c48ce0db3b058d8c1cea2f39da13277a4536973eeb87a3f9eb32bab7feea4126

Prebuilt hg38 reference indexes v1 (reference data only)

Prebuilt hg38 reference indexes — v1

Reference data only: this does not replace the latest OncoTracer software release.
The bundle contains the public UCSC hg38 genome from AWS iGenomes, plus BWA and minimap2 indexes. It contains no patient reads, sample information, tumor predictions, or model weights.

Install the command

The existing v2.0.0 executable does not have reference install. Use this tested source revision in a new software directory:

git clone --branch hg38-reference-v1 https://github.com/cfarkas/oncotracer.git oncotracer-src
python3 -m venv oncotracer-env
oncotracer-env/bin/python -m pip install -e ./oncotracer-src
source oncotracer-env/bin/activate

Keep both folders. Linux and Python 3.10–3.13 are required. The reference tag fixes source revision 6419b2aefba915457ff98da660ffbc4d7fc8f414; Git's detached-HEAD notice is expected for a tagged checkout.

Download and install the indexes

Choose a new, reusable reference parent, replacing /absolute/path/shared-reference below:

oncotracer reference install \
  --manifest https://github.com/cfarkas/oncotracer/releases/download/hg38-reference-v1/hg38-reference.json \
  --sha256 1c704b1522fe37c13bf4272858792d03e827bc476fdb6712fa7f760f3c944632 \
  --lpwgs-root /absolute/path/shared-reference \
  --mode ont --dry-run

--manifest is the inventory URL; --sha256 verifies it; --lpwgs-root is where reusable references will live. --mode ont installs the ONT index, illumina selects BWA, and both installs both. --dry-run previews space and destination. Remove it to download and install. You do not need to download individual .part files manually.

Set lpwgs_root: /absolute/path/shared-reference in the project YAML. The installed directory is shared-reference/references/samurai_hg38/. Existing references are never overwritten. An interrupted import can be restarted, but partial downloads are not resumed.

Allow approximately 8.0 GiB for Illumina, 9.7 GiB for ONT, or 14.8 GiB for both, plus 1 GiB of free headroom. The importer verifies each chunk and complete file without loading or building an index. The complete public HTTPS download/import passed under a 1-GiB address-space limit with 42,940 KiB peak RSS (about 42 MiB), no swap and zero index builds. This is not an alignment-memory requirement: analysis needs more RAM. Use oncotracer system --path /absolute/path/project for planning guidance.

Tool compatibility

These indexes were validated with Linux x86-64 Bioconda BWA 0.7.19=h577a1d6_1 and minimap2 2.30=h577a1d6_0. Analysis verifies exact executable hashes and stable package metadata; a different build is rejected rather than silently rebuilding the shared reference. Other architectures/builds need a separately validated bundle.

Fresh installations of both tools accepted the public download through OncoTracer's normal reference checks and correctly mapped synthetic short/long reads without rebuilding indexes. This checks index usability, not a complete CNA workflow or tumor-classifier accuracy.

The manifest pins all genome, index, and chunk hashes. Usability changes and software test status are tracked separately. This is research software, not a standalone diagnostic system.

OncoTracer v2.0.0

OncoTracer v2.0.0

Native, copied-executable OncoTracer release with no Nextflow dependency in the v2 runtime.

Validated on exact commit: b5ce9fc01c88e69935a53dd7c36ed3c608d56b54

• Native v2 CI passed on the exact release commit.
• QuickStart 1 passed semantic v1.1/native-v2 parity for Illumina and ONT.
• QuickStart 2 passed semantic v1.1/native-v2 parity for all three HCC1143 libraries.
• The frozen comparator uses OncoTracer v1.1, SAMURAI v1.4.0, pinned Nextflow, and immutable container digests.
• Container: ghcr.io/cfarkas/oncotracer@sha256:ea8b4e771134d8b2c84afd3e457917009ac674cd305de415922adf2322940d14

OncoTracer v1.1

OncoTracer v1.1

Stable research release with four clearly documented launch methods:

1. Nextflow with Docker (--docker)
2. Nextflow with Singularity or Apptainer (--singularity)
3. The Poetry-managed launcher (poetry run oncotracer)
4. Nextflow with native Conda environments (--conda)

QuickStart 1 and QuickStart 2 passed complete public-data validation through Docker, Poetry with Docker, and Conda on scientific-code commit 0f1a3d86521afcb5eac3ce6e65fd45d455e8c843. Final release commit 032c1268fa7fdcadc48087055066d7a9fc59bd89 adds documentation and documentation-regression coverage only; it does not change the scientific workflow code.

The documentation gives explicit Docker, Singularity/Apptainer, Poetry, and Conda commands in QuickStart 1, QuickStart 2, Automatic Setup, the complete PRJNA754199 tutorial, and the mock six-tumor/four-normal example.

Validation covered public-read integrity, Automatic Setup, SAMURAI, qDNAseq/ichorCNA, BAM refinement, CNA tables, plots, workflow summaries, BAM integrity, final output verifiers, portable Docker ownership, the Poetry launcher, native Conda environments, and the pinned local SAMURAI source cache.

Singularity/Apptainer uses the same maintained container image as Docker and is documented for configured HPC systems.

OncoTracer is for research use and is not a standalone diagnostic system.

OncoTracer v1.0

OncoTracer v1.0

First stable research release of OncoTracer.

Validated on commit e0e7f944c207b830ffef8f2dcdf3869b536b7e05 with the public ONT QuickStart 1 sample DRR165691 using Nextflow and Docker. The validation completed read preparation, minimap2 alignment, SAMURAI/ichorCNA, BAM-based boundary refinement, CNA tables, cytogenomic notation, plots, and the workflow summary.

The public three-library HCC1143 Illumina QuickStart 2 had also completed successfully before this release.

OncoTracer is for research use and is not a standalone diagnostic system.