checking

SigProfiler Suite · de novo mutational signatures

Find the mutational signatures shaping your samples

Every process that damages DNA, from tobacco smoke and ultraviolet light to failing DNA repair, leaves its own pattern of mutations. SigProfilerExtractor separates a cohort's mutations into those patterns, mutational signatures, and reports how much each one contributed to every sample. It tries every plausible number of signatures, keeps the most stable solution and matches it to the COSMIC catalogue of known signatures.

This Workbench runs it on your own computer, from variant calls or a ready-made matrix to an interactive report, without writing code.

The example finishes in about three minutes on a laptop.

You give

one cohort per run
  • Variant callsA folder of VCF files, one per genome (MAF and simple text work too). Matrices are built for you by SigProfilerMatrixGenerator.
  • A mutation matrixMutation types down the rows, samples across: SBS96 to SBS4608, DBS78, ID83, CNV48 or SV32.
  • Structural variantsA folder of BEDPE files, classified into the 32 SV channels.
  • Copy-number segmentsOne segmentation file from ASCAT, Battenberg, FACETS, PURPLE, Sequenza and others, classified into 48 CNV channels.

You get

for every mutation type
  • How many signaturesThe suggested number, chosen from stability and reconstruction error across every number tried.
  • De novo signatures and activitiesEach signature's profile, and how many mutations it contributed to each sample.
  • COSMIC decompositionEach de novo signature explained by known COSMIC signatures (v3.6 by default), with sample activities.
  • Mutation probabilitiesWhich signature most likely caused each mutation type in each sample.
  • Plots and an interactive dashboardPDFs for every solution, and a self-contained report to explore and share.

How it works

the method of Islam et al., Cell Genomics 2022
  1. 1MatricesVariant calls are counted into mutation matrices (SBS96, DBS78, ID83 and more), one column per sample.
  2. 2Many factorizationsFor each candidate number of signatures, non-negative matrix factorization runs on many resampled copies of the data: 100 by default.
  3. 3Stable signaturesThe replicate solutions are clustered. A signature found again and again across replicates is stable.
  4. 4The number of signaturesThe suggested solution is the largest stable one that still reconstructs the samples well.
  5. 5Match to COSMICSigProfilerAssignment explains each de novo signature with known COSMIC signatures and assigns them to every sample.
BenchmarkedCompared with 13 other tools on more than 60,000 simulated cancer genomes, it found more of the true signatures and far fewer false ones.
ReproducibleEvery run records its seed in Seeds.txt. Give it back to repeat the run exactly.
PrivateEverything runs on this computer. Your data never leaves it.
Same results as codeThe Workbench calls the same function as Python and the command line, and writes the same files.

Quit the Workbench?

This stops the app. Your results stay where they are.

Delete this run for good?