Skip to content

After ONT: drag samples into a CpG classifier

Start methylation directly from named ONT inputs, or reuse a saved ONT project. A completed CNA analysis is not required. Keep original inputs grouped by intended library, using the same names as the ONT setup:

ONT/
├── ONCO-001/
│   ├── fastq_pass/
│   ├── fastq_fail/
│   ├── pod5_pass/
│   ├── pod5_fail/
│   └── bam_pass/       # or modified_bam/
└── ONCO-002/
    ├── fastq_pass/
    ├── fastq_fail/
    ├── pod5_pass/
    ├── pod5_fail/
    └── bam_pass/       # or modified_bam/

Methylation automatically links each named library's pod5_pass. The browser lists its ONCO-style name; drag samples to MARLIN or Sturgeon using clinical records. Saved or explicitly selected modified-base BAMs remain valid alternatives.

CNA/variants use only fastq_pass. FASTQs define sample read membership but contain no methylation marks. CpG analysis also needs matching original modified-base BAMs with MM/ML tags, or POD5 plus compatible Dorado models. The fail folders are retained originals, not additional analysis inputs; signal folders are optional for CNA/variants. Legacy MinKNOW/barcode and single ligation layouts remain supported. See ONT input settings.

New project from named inputs: open ordinary setup, discover the libraries, and assign samples to the appropriate classifier group:

oncotracer --interactive --backend conda --mode ont --analysis methylation --input-folder /data/ONT

Reuse a saved ONT project: restore its names, roles, reference and reusable settings directly into the CpG assignment page. The original stays unchanged:

oncotracer --interactive --cpg-from /path/to/ONT-project

These named-input and reuse routes require the current main source; use the GitHub installation option. Previously published releases and PyPI 2.1.0 do not include it.

Open the playable guide.

Watch: restore, drag and save

Transcript · Download video

Actual browser controls, synthetic named libraries and POD5 files. The video shows the saved-project reuse route, pointer drags including a return to Unknown, and separate checks. No analysis or patient data. Fixture models are not scientific models.

1. Drag samples

Drag documented leukemia samples to MARLIN, or CNS-tumour samples to Sturgeon. Use the dropdown if preferred. Choose from clinical records, not a previous classifier prediction. Cancer/Normal records the clinical role; MARLIN/Sturgeon/Unknown chooses a CpG classifier independently. Drag unknown or unsuitable samples to Unknown / do not classify. This excludes their CpG classification without changing their clinical role or labeling them normal.

2. Save and check

Choose the classifier group, then Save configuration and check. OncoTracer suggests a separate ONT-project_MARLIN or ONT-project_Sturgeon folder. Only the chosen group is included; other samples are not queued.

If a requirement is missing, the page shows what to supply once: original modified-base BAMs or POD5, installed classifier resources, and any required license confirmation. For Sturgeon, read the linked license and explicitly acknowledge its terms, then choose Download / verify Sturgeon general model. This fills the pretrained model and hg38 probe paths after verification; your Sturgeon executable must also be installed. Resolve missing items before running. See model acquisition and recorded acknowledgement.

The video demonstrates assignment and Save/check with fixtures. The model-download control described here is available in the current GitHub version.

3. Run when ready

Click Run analysis explicitly after checks pass. The video stops after saving. To run the other group, open the same command in another terminal and choose that group. You can append --classifier marlin or --classifier sturgeon to preselect the group; still review its sample assignments.

Input requirements and result details

From the current source checkout, the equivalent Nextflow entry is:

nextflow run main.nf --interactive --cpg_from /path/to/ONT-project --backend conda
  • The original sample FASTQs select read membership. Modified-base BAMs must contain MM/ML tags; an ordinary CNA BAM is insufficient. Alternatively, completed POD5 files need compatible Dorado basecalling/modification models.
  • Reused tool/model paths must exist. Optional methylation tools and licensed assets are separate from the base installation; use the resource reference. Setup does not download them or accept terms for you. Native Conda/host/Poetry is required.
  • Each project runs one classifier. Divide CPU/GPU capacity between concurrent projects; see run, stop and resume.
  • Review results/07_methylation/methylation_status.json, classifier scores and usable CpG/probe coverage. Missing predictions do not mean normal. See result meanings.