Vocabulary
The words run() accepts for tumor_type, seq_type, soi and catalog, and the words it uses in its output. Names match in any case; the tables are generated from the installed release (SigMA2 0.0.16-alpha.20, generated 2026-11-05).
Terms
| Word | Meaning |
|---|---|
SigMA2, sigma2 | the product, and the package |
platform, seq_type | the assay a sample was sequenced with, named by its capture design |
| classifier | what runs: one cell of tumor type, signature and platform |
| model | the stored files of a classifier |
signature of interest, soi | the signature a run scores |
| cell | one tumor type, one signature, one platform |
| card | what a classifier records about itself (metadata.json) |
| floor | the SNV count below which a sample is not scored |
| operating point | a cutoff on the score set at a false-positive rate: loose (10%) or strict (1%) |
| store | a directory of models in the <tumor_type>/<soi>/<seq_type> layout |
| Add object | the training object: genomes, exposures, labels, clusters, simulations, models |
Tumor types
tumor_type takes one of the names below; the package's list holds 39, msi and pole among them, which are signature sets. The short ones: prost (prostate), crc (colorectal), lymph (lymphoma), eso (esophagus), head (head and neck), lgg (low-grade glioma), panc_ad (pancreatic adenocarcinoma), panc_en (pancreatic neuroendocrine), medullo (medulloblastoma), osteo (osteosarcoma), lms (leiomyosarcoma), meso (mesothelioma), bone_other. Endometrial carcinoma is uterus.
A tumor type sets the decomposition's signature set, the cluster centroids and the classifier. A type with no signature set in the catalog cannot be scored and fails early. msi and pole are the signature sets the decomposition basis is extended with, not tumor types to pass.
| tumor_type | SNV floor | signature set in cosmic_v3p2_inhouse |
|---|---|---|
| adrenal | 5 | yes |
| biliary | 5 | yes |
| bladder | 5 | yes |
| bone_other | 5 | yes |
| breast | 5 | yes |
| cervix | 5 | yes |
| crc | 5 | yes |
| eso | 5 | yes |
| ewing | 5 | yes |
| glioma | 5 | yes |
| head | 5 | yes |
| kidney | 5 | yes |
| leuk | 5 | yes |
| lgg | 5 | yes |
| liposarc | 5 | yes |
| liver | 5 | yes |
| lms | 5 | yes |
| lung | 5 | yes |
| lymph | 5 | yes |
| medullo | 5 | yes |
| meso | 5 | yes |
| msi | 5 | yes |
| myeloma | 5 | yes |
| osteo | 3 | yes |
| ovary | 5 | yes |
| panc_ad | 5 | yes |
| panc_en | 3 | yes |
| paraganglioma | 5 | no |
| pole | 5 | yes |
| prost | 4 | yes |
| sarc | 5 | no |
| sarc_other | 5 | no |
| skin | 5 | yes |
| stomach | 5 | yes |
| testicular | 5 | no |
| thymoma | 5 | no |
| thyroid | 5 | yes |
| uterus | 5 | yes |
| uveal | 5 | no |
Platforms
seq_type takes one of these names. Six of them have classifiers; the rest are capture designs you can simulate panels for in training when you supply their BED.
| seq_type | signature classifiers | HRD, MMRD/POLE | SigMA v1 (legacy) | bundled regions |
|---|---|---|---|---|
| msk | yes | yes | yes | yes (hg19) |
| seqcap | yes | yes | yes | yes (hg19) |
| wgs | yes | — | yes | — |
| seqcap_probe | — | — | yes | — |
| fo | yes | yes | — | yes (hg19) |
| msk_on_target | — | — | — | — |
| ampliseq | — | — | — | — |
| illu_hotspot | — | — | — | — |
| illu_trusight | — | — | — | — |
| invitae_breast | — | — | — | — |
| invitae_cns | — | — | — | — |
| invitae_crcp | — | — | — | — |
| invitae_cr | — | — | — | — |
| invitae_gastric | — | — | — | — |
| invitae_gyn | — | — | — | — |
| invitae_multi | — | — | — | — |
| invitae_panc | — | — | — | — |
| invitae_renal | — | — | — | — |
| tcga_mc3 | yes | yes | yes | yes (hg19) |
| wgs_pancan | — | — | yes | — |
| op | yes | yes | — | yes (hg19) |
seq_type="auto" chooses among the platforms with a classifier by SNV burden and reports the choice. A custom panel name is not accepted.
Signatures
soi takes a COSMIC SBS name (SBS3, SBS8, SBS18, SBS7a, …), one of the composite presets, an alias, or HRD.
| Composite | Parts |
|---|---|
APOBEC | SBS2, SBS13 |
SBS7 | SBS7a, SBS7b, SBS7c, SBS7d |
SBS10 | SBS10a, SBS10b, SBS10c, SBS10d |
SBS17 | SBS17a, SBS17b |
CLOCK | SBS1, SBS5 |
Aliases: UV is SBS7, POLE is SBS10. A composite is scored by one classifier trained for it; its features are named by the composite (APOBEC_mva, exp_APOBEC). A custom composite has no classifier until you train one.
Catalogs
catalog takes one of these. Shipped SigMA2 classifiers use cosmic_v3p2_inhouse; SigMA v1's classifiers and MMRD/POLE use cosmic_v2_inhouse. Leave it out and the classifier's own is used. "inhouse" is SigMA's curated copy of COSMIC with signature sets per tumor type.
| catalog | tumor types with a signature set |
|---|---|
| cosmic_v2_inhouse | 35 |
| cosmic_v3_inhouse | 36 |
| cosmic_v3p2 | 32 |
| cosmic_v3p2_inhouse | 37 |
Status words
| Where | Words |
|---|---|
Card.status, describe() | released (●), weak (◐, released with weak evidence), experimental (▲) |
call_evidence | informative, weak |
confidence | high, low, negative, NA |
hrd_call | HRD, HRD-low, HRP, MMRD, POLE |
hrd_group | a tumor type, or pooled |
hrd_variant | gis, tcn, nogis |
MMRD_category | MMRD, POLE, MMRP; Unclassified from the legacy model only |
snv_cutoff_source on a card | the training summary, the tune result, or the tumor type's default |
Column naming
| Pattern | Example | Meaning |
|---|---|---|
<soi>_mva | SBS3_mva | the classifier's score |
exp_<soi>, rat_<soi> | exp_SBS3, rat_SBS3 | exposure and exposure fraction |
<sig>_c, <sig>_l_rat, <sig>_wl, <sig>_c_diff | SBS8_c | per-signature features of the decomposition |
<soi>_high_<k>_ml, <soi>_low_<k>_ml | SBS3_high_1_ml | likelihood of the k-th cluster with and without the signature |
<soi>_pass_mva, <soi>_confidence | SBS3_pass_mva | in an HRD run, the Signature 3 step's own calls |
a[c>a]a | a channel, the package's spelling of COSMIC's A[C>A]A |
Full glossary: Read the output.