ONT methylation: leukemia or CNS classification¶
Choose MARLIN for leukemia or Sturgeon for CNS-tumor research. Select the classifier explicitly; review predictions alongside other laboratory findings.
Install first. Browser and terminal routes are below; see resources and SSH/headless for details.
Finished an ONT copy-number run? Follow the post-ONT CpG video guide to reuse original inputs and assign separate MARLIN/Sturgeon projects.
What you need¶
You need matching FASTQs, methylation input, and an installed classifier:
| Input | Why it is needed |
|---|---|
| FASTQs in named libraries or legacy barcodes | Select each sample’s read IDs |
| Modified-base BAMs or raw POD5 files | Supply methylation information |
| Dorado, Modkit, and samtools | Align reads and extract methylation |
| MARLIN or Sturgeon tools and model files | Compare the methylation pattern with known classes |
Modified-base BAMs contain MM/ML tags; reusing them avoids basecalling. Ordinary BAMs or FASTQs alone are insufficient. Raw POD5 needs compatible Dorado basecalling and modification models.
Use completed files; exclude live files, duplicates and alternate basecalls. Keep one intended library per named folder. Select legacy barcodes explicitly; include unclassified only with a justified assignment, never pooled across patients.
1. Prepare the tools once¶
Use Conda. Methylation tools and classifier assets are separate from oncotracer install --conda; see local resources.
Locate executables, model and probe BED; MARLIN also needs feature-order .RData and class-annotation .xlsx files. Setup records checksums. Probes are genomic sites recognized by the classifier.
2. Link inputs in the browser¶
oncotracer setup
Choose Oxford Nanopore, then Fresh or FFPE from specimen records. Select
named library folders, or Browse run for a legacy MinKNOW run.
Named libraries automatically link pod5_pass; saved/explicit modified BAMs stay
selected. Failed siblings are excluded. Legacy barcode/ligation layouts also work.
Assign and name the samples, then choose Methylation classification or CNA and methylation.
Choose the linked POD5 or modified-base BAM input. For methylation-only analysis, drag included samples into the MARLIN or Sturgeon classifier group according to documented clinical context, then select Classifier group to save and run. Other groups and unassigned samples are excluded, not queued; save each classifier to a separate project. CNA and methylation keeps one chosen classifier for all selected samples.
Browse to the installed tools and model files, or reuse a resource YAML. Modkit extracts CpG methylation for either classifier. Selected libraries and their FASTQ read IDs determine which signal reads belong to each sample. Save, check, and click Run analysis for the last saved project.
Terminal example: leukemia using existing BAMs¶
Replace the paths, barcode, and sample name below with yours:
oncotracer setup --terminal \
--project /work/leukemia-study --backend conda \
--mode ont --analysis methylation \
--reads-folder /data/run/fastq_pass \
--barcodes barcode01 --sample-names sampleA \
--classifier marlin \
--modbam /data/run/bam_pass \
--cpu --threads 8
This asks for remaining tool/model paths in the terminal. Check/run the saved
/work/leukemia-study/config/run.yml below. --no-browser still starts a server.
For unattended reuse, use a complete resource YAML from a previous project. Missing tools or assets stop setup. This alternative checks and runs automatically:
oncotracer setup --non-interactive --run \
--project /work/new-leukemia-study --backend conda \
--resources /work/previous-study/config/run.yml \
--mode ont --analysis methylation --classifier marlin \
--reads-folder /data/new-run/fastq_pass \
--barcodes barcode01 --sample-names sampleB \
--modbam /data/new-run/bam_pass --cpu --threads 8
| Flag | Meaning |
|---|---|
--analysis methylation |
Run methylation only; use both to also request copy-number analysis |
--reads-folder |
Parent of the selected barcode FASTQ folder |
--barcodes / --sample-names |
Which barcode to use and the name for its results |
--classifier marlin |
Use the leukemia classifier |
--modbam |
Existing BAM file or directory, aligned or unaligned; calls are reused and reads aligned to hg38 on CPU |
--cpu |
Keep methylation tools on CPU, including MARLIN |
--threads 8 |
Request eight CPU worker threads |
For CNS, the explicit terminal counterpart is below. Confirm the applicable Sturgeon license when prompted; setup does not grant a license.
Terminal / headless: CNS using existing BAMs¶
oncotracer setup --terminal --run --project /work/cns-study \
--backend conda --mode ont --analysis methylation \
--reads-folder /data/run/fastq_pass --barcodes barcode01 --sample-names sampleA \
--classifier sturgeon --modbam /data/run/bam_pass --cpu --threads 8
If you only have raw POD5¶
These complete terminal alternatives ask for the classifier resources plus matching Dorado basecalling/5mCG/5hmCG models, then check and run. FASTQs still define sample membership. Use a new project for each alternative.
Terminal / headless: leukemia using POD5¶
oncotracer setup --terminal --run --project /work/leukemia-pod5-study \
--backend conda --mode ont --analysis methylation \
--reads-folder /data/run/fastq_pass --barcodes barcode01 --sample-names sampleA \
--classifier marlin --pod5-dir /data/run/pod5_pass --cpu --threads 8
Terminal / headless: CNS using POD5¶
oncotracer setup --terminal --run --project /work/cns-pod5-study \
--backend conda --mode ont --analysis methylation \
--reads-folder /data/run/fastq_pass --barcodes barcode01 --sample-names sampleA \
--classifier sturgeon --pod5-dir /data/run/pod5_pass --cpu --threads 8
CPU basecalling can be slow for large inputs. --gpu enables GPU basecalling/MARLIN; keep --cpu when the GPU is busy. Modkit and Sturgeon use CPU.
3. Check and run¶
For the saved leukemia-BAM example above (the --run alternatives already execute):
oncotracer check --config /work/leukemia-study/config/run.yml
oncotracer run --backend conda \
--config /work/leukemia-study/config/run.yml --cpu
check verifies paths/settings, not biological quality. Run checks MARLIN dependencies before processing and may download hg38 into the project reference cache.
Reuse tool/model settings with setup --resources /work/leukemia-study/config/run.yml; supply new sample paths. See oncotracer setup --help for resource flags.
Read the result¶
Within your project, open results/07_methylation/methylation_status.json first. Each sample has a status and paths to any outputs:
| Status | What it means | Next step |
|---|---|---|
complete |
The requested classifier produced output | Review scores and data quality; completion alone does not make the call reliable |
no_cpg_modifications |
No usable modified-CpG calls were found | Check the modified-base calls, human alignment, and read yield |
no_classifier_probes |
CpG calls exist, but none overlap the supplied MARLIN probes | Check hg38 coordinates and usable coverage; no leukemia prediction was made |
failed |
A tool, resource, or input check failed | Read the error and the logs in 07_methylation/logs/ |
For MARLIN, inspect covered_classifier_probes. Total read count alone does not establish sufficient classifier coverage. A missing prediction is not evidence that the sample is normal.
If you requested both branches, the summary records methylation and copy-number outcomes separately. Successful outputs are kept even when another requested branch fails. Detailed processing and resource reference.