Try it: read a fingerprint, then blend two causes
The charts below are drawn in this page by mSigSDK’s own COSMIC SBS96 renderer, the same component the tutorial site uses, from the real COSMIC Mutational Signatures v3.6 profiles. Pick a cause to read its 96-channel fingerprint, then blend two causes and watch the fingerprint change.
Data source shipped on this page: the COSMIC Mutational Signatures v3.6 catalog (GRCh37 SBS96 file, downloaded from cancer.sanger.ac.uk), profiles SBS1, SBS4, and SBS7a, values exactly as published. The SDK’s bundled demo data ships only small teaching spectra, so this page uses the catalog file directly. COSMIC signature profiles are © COSMIC, Wellcome Sanger Institute: free for research and academic use with attribution; commercial use requires a separate license from COSMIC. Renderer and SDK code: mSigSDK, MIT license, copied from episphere.github.io/msig.
Explore: 96-channel signatures
Loading the SDK renderer and the COSMIC v3.6 profiles…
Aging — clock-like C>T changes that accumulate in stem cells over a lifetime; found in nearly every cancer.
Both charts share one fixed vertical scale: percent of mutations, 0% to 36.5% (36.5% is the tallest bar in these profiles, SBS1 at A[C>T]G). The scale never changes when you switch causes or blend, so bar heights can be compared across both charts. Each chart is stacked by substitution class so all 96 trinucleotide contexts stay labeled at any screen width.
Mixture: blend two causes
Real cancers are mixtures. Choose both causes, then set the share. The blend redraws live.
Loading the SDK renderer and the COSMIC v3.6 profiles…
Blend shown: 60% Aging (SBS1) and 40% Smoking (SBS4). The two shares always add to 100%.
Browser-native computation of mutational signatures: four cancer-signature tools, one browser tab, zero installs, and a clear view of where they disagree.
mSigSDK runs four established signature tools side by side in the browser, with no install and no data leaving the machine: SigProfilerAssignment, MuSiCal, deconstructSigs, and sigminer, through one adapter layer. It adds its own NNLS fitting and schema-validated portable reports.
The disagreement is the finding
On 2,700 synthetic spectra with known truth, every tool’s exposure estimates stayed close (mean cosine above 0.96), but at 10% noise their calls for which signatures are active ranged from F1 0.532 to 0.924. On the real PCAWG cohort the four tools called between 4.13 and 17.39 active signatures per sample. Each adapter reproduced its package’s own results on 38 public PCAWG lung adenocarcinoma SBS96 spectra. Paper: DOI 10.1186/s12859-026-06681-z, BMC Bioinformatics 2026. Site: episphere.github.io/msig. Code (MIT): github.com/episphere/msig.