Skin cancer methylation classifier
Methylation-based classification of skin cancers.
How it works
Methylation-based classification of skin cancers
Preliminary version. This classifier is in development. The specifications below describe the current training iteration and may change before release. It is not yet available for licensing or routine diagnostic use.
The classifier assigns skin and cutaneous soft-tissue tumours to a methylation class using the same approach as our CNS and sarcoma classifiers: a machine-learning model trained on an annotated reference cohort, returning a confidence score for every class.
An expanding reference cohort
The current iteration is trained on 1,588 methylation profiles covering 51 distinct methylation classes, including control classes for dermis and whole blood leukocytes so that non-tumour material is recognised as such.
A calibrated confidence score
Every class receives a score between 0 and 1, so borderline cases are visible as borderline. Cutaneous entities overlap considerably on morphology, which is exactly where an objective molecular score adds most.
Broad coverage of cutaneous entities
Classes span basal cell carcinoma and cutaneous squamous cell carcinoma, melanoma and its variants, Merkel cell carcinoma, mycosis fungoides, Langerhans cell histiocytosis, dermatofibrosarcoma protuberans, angiosarcoma and Kaposi sarcoma, among others.
Single-tier output
Results are returned as a methylation class, alongside a copy number profile and quality-control metrics.
Specifications
Current training iteration
Coverage
- 51 distinct methylation classes
- Training set of 1,588 methylation profiles
- Includes control tissue classes
Performance
- Single-tier output structure
- Proposed evidence level annotation
- Confidence score for every call
Status
- Preliminary: in development
- Not yet available for routine diagnostic use
- Specifications may change before release
- Early evaluation possible on request
Workflow
From sample to report
- Sample collection: fresh frozen or FFPE, tumour content at least 70%
- DNA extraction
- Raw data generation
- Classification
- PDF and HTML report
Every report includes
- Quality control metrics
- Tumour classification with confidence score
- Copy number variation analysis
Supported platforms
Platform-agnostic: now compatible with sequencing data
Like our other classifiers, this model is not tied to one instrument. A single model handles both array and sequencing input, so a laboratory can add or change technology without changing classifier.
Methylation arrays
- Illumina Infinium MethylationEPIC v2
- Illumina Infinium MethylationEPIC v1
- Illumina HumanMethylation450 (450k)
Upload unprocessed IDAT files and the sample is classified automatically.
Sequencing
- Nanopore whole-genome sequencing
- Nanopore adaptive sampling
- Whole-genome bisulfite sequencing (WGBS)
- EM-seq
- TAPS+
- Illumina 5-base
- biomodal 6-base
- Targeted methylation panels
Sequencing-derived methylation data is classified with the same model, with no ad-hoc per-platform variant.
Early access
Interested in the skin cancer classifier?
The skin cancer methylation classifier is still in development and is not yet licensed for routine diagnostic use. We work with clinical and research partners during development, and there is scope for early evaluation.
If you would like to be notified on release, or to discuss evaluating the current version, get in touch at support@epignostix.com. For licensing questions about our released classifiers, contact licensing@epignostix.com.
