SigMA2
SigMA2 classifies tumors from the somatic mutations of a targeted gene panel or an exome, where the mutation count is far below what whole-genome signature analysis needs. It makes three predictions per tumor: whether its point mutations carry a given mutational signature; whether its mutations and copy number resemble those of tumors with biallelic BRCA1/2 inactivation (HRD); and whether its mutations resemble those of microsatellite-unstable or POLE-mutated tumors (MMRD, POLE). Each is a prediction from sequencing data, not a functional or clinical determination; what a call means states what each supports.
The same runs are available from the command line for HRD (Command line) and in sigscape studio, where the table is rendered and every run is kept.
First steps
- Choose the analysis and check your panel: which analysis answers your question, and whether your files and panel can support it.
- Install: the package and a reference genome, about ten minutes.
- Quickstart: fifty public breast panels scored for Signature 3, APOBEC and mismatch-repair deficiency, about five minutes.
- The guide for your goal, below.
Coming from SigMA in R: Migrate from SigMA v1 runs your first analysis with SigMA v1's classifier and translates the arguments, columns and functions.
Goals
Which analysis answers my question, and can my panel support it?
- You need
- A description of your data: file format, panel, tumor type, genome build.
- You get
- The analysis to run and the inputs it needs.
- Next
- Install, Quickstart
Does a tumor carry a mutational signature?
- You need
- The somatic mutations of your panels or exomes (MAF or VCFs), the tumor type, the platform.
- You get
- A score, a call at a 10% and at a 1% false-positive rate, and a confidence word per tumor.
Is a tumor homologous-recombination deficient?
- You need
- Mutations with indels, the repeat-locus files (from the maintainers or the public SigMA repository), and copy-number segments when you have them.
- Start at
- 3. Detect HRD
- You get
- A prediction per tumor, HRD, HRD-low or HRP, with the Signature 3 call beside it.
Is mismatch repair deficient, or POLE mutated?
- You need
- Mutations with indels and the repeat-locus files, from the maintainers or the public SigMA repository.
- Start at
- 4. Detect MMRD or POLE
- You get
- MMRD, POLE or MMRP per tumor.
Move from SigMA in R to SigMA2
- You need
- The package installed; your R scripts.
- Start at
- Migrate from SigMA v1
- You get
- A first SigMA2 run with SigMA v1's classifier, and the arguments, columns and functions translated.
Run SigMA2 on data from my own pipeline
- You need
- The package installed; your mutations as files or tables.
- Start at
- Adapt SigMA2 to your own data
- You get
- Inputs SigMA2 accepts, a run in your pipeline, and what to record beside each result.
Match a result from SigMA in R
- You need
- The same input, and the settings R used.
- Start at
- Reproduce SigMA v1
- You get
- SigMA v1's scores and calls, reproduced in Python.
Train a classifier for a tumor type, signature or panel the package does not cover
- You need
- Whole genomes with known exposures, sigma2[panel] and bedtools.
- Start at
- Train a classifier
- You get
- A classifier in your own store, graded by the release gate.
The guides
The guides are in three groups: Getting started covers choosing the analysis, the first install, the first result and the move from SigMA v1; Core guides cover scoring and reading results; Advanced guides cover your own data and pipelines, training, curation, the store and SigMA v1's classifiers. Select a guide to see who it is for, what it needs, what it produces, and what to read next.
- For
- Anyone who wants a first result on public data.
- Teaches
- How to score Signature 3, APOBEC and mismatch-repair deficiency on fifty breast panels and read the result table.
- Needs
- the package installed; the hg19 genome; the repeat-locus files for step 6.
- Result
- A scored table with calls and confidence words.
How far to trust a call
Every classifier in SigMA2 was tested on tumors it had not seen, and each has a status: released (a positive call holds at the stated false-positive rate), released with weak evidence (the false-positive rate holds, but a positive call changes the odds little), or experimental (scores for ranking a cohort, no calls). 39 breast, lung and skin cells were released before the gate and carry no evidence grade, and the HRD and MMRD/POLE models are cross-validated without a gate. Release status explains the three; the bundled classifiers page lists the status of every tumor type, signature and platform. SigMA2's documentation quotes no sensitivity, specificity or likelihood-ratio figures; the status states what a classifier's calls support.
Reference
Python API, Command line, Vocabulary, Files and formats, Example files and environment, For AI agents. The concept pages explain how a classifier scores a panel, floor, cutoffs and catalog and the known limits.
These pages describe SigMA2 0.0.16-alpha.20, the release on PyPI. The run() arguments and the output columns are the stable surface.