Assigning mutational signatures to individual samples and individual somatic mutations with SigProfilerAssignment
Marcos Díaz‐GayiD, Raviteja VangaraiD, Mark D. BarnesiD, Xi WangiD, S. M. Ashiqul IslamiD, Ian VermesiD, Stephen Oscar DukeiD, Nithish Bharadhwaj Narasimman, Ting YangiD, Zichen Jiang, Sarah MoodyiD, Sergey SenkiniD, Paul BrennaniD, Michael R. StrattoniD, Ludmil B. AlexandroviD
Abstract
Analysis of mutational signatures is a powerful approach for understanding the mutagenic processes that have shaped the evolution of a cancer genome. Here we present SigProfilerAssignment, a desktop and an online computational framework for assigning all types of mutational signatures to individual samples. SigProfilerAssignment is the first tool that allows both analysis of copy-number signatures and probabilistic assignment of signatures to individual somatic mutations. As its computational engine, the tool uses a custom implementation of the forward stagewise algorithm for sparse regression and nonnegative least squares for numerical optimization. Analysis of 2,700 synthetic cancer genomes with and without noise demonstrates that SigProfilerAssignment outperforms four commonly used approaches for assigning mutational signatures. SigProfilerAssignment is freely available at https://github.com/AlexandrovLab/SigProfilerAssignment with a web implementation at https://cancer.sanger.ac.uk/signatures/assignment/.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 194
- FWCI
- 24.60
- Field percentile
- top 1%
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (11 Jul 2023) · CC BY 4.0
SigProfilerAssignment refits known substitution, doublet, indel and copy-number mutational signatures to single samples and probabilistically to individual mutations, outperforming four refitting tools on 2,700 synthetic genomes.
Existing refitting tools (deconstructSigs, MutationalPatterns, sigLASSO, SignatureToolsLib) mostly lack web interfaces, focus on substitution signatures, were rarely benchmarked against each other, and none assigns signatures to individual mutations. SigProfilerAssignment offers desktop and online refitting for all COSMIC signature classes, including copy-number, adds per-mutation probabilities, and benchmarks itself against these tools.
Key results
ShownIn synthetic benchmarks of single base substitution (SBS) refitting, SigProfilerAssignment was more accurate than deconstructSigs, MutationalPatterns, sigLASSO and SignatureToolsLib at every tested noise level. It was the only tool to keep F1 above 0.90 at the highest noise level.
2,700 simulated genomes (300 tumors from nine cancer types) built from 21 COSMIC signatures and refit against all 79 COSMICv3.3 SBS signatures, with 0%, 5% and 10% random noise; at 10% noise only SigProfilerAssignment reached F1 >0.90; precision, sensitivity and F1 were computed per sample and averaged. · MAIN; Benchmarking of bioinformatics tools for…; Figure legends
SuggestedSigProfilerAssignment combined high precision with better sensitivity than the comparators. The MutationalPatterns standard fit lost precision at all noise levels, which the authors attribute to overfitting; its strict mode reduces this but runs much more slowly.
Qualitative readout from precision versus sensitivity plots at 0%, 5% and 10% noise; MutationalPatterns v3.0.1 standard (fit_to_signatures) vs strict (fit_to_signatures_strict, max_delta 0.004). Other tools limit overfitting with an L1 penalty (sigLASSO), a sum-squared-error penalty plus a mutation-fraction filter (deconstructSigs) or a mutation-fraction filter (SignatureToolsLib). · MAIN; Benchmarking of bioinformatics tools for…; Figure legends
ShownSigProfilerAssignment ran quickly with low memory use. Only the standard mode of MutationalPatterns was substantially faster, and none of the benchmarked tools needed significant memory.
SigProfilerAssignment processed 2,700 samples in 9.6 minutes (0.21 seconds per sample); CPU elapsed time and peak memory were monitored and averaged across the three noise levels; the faster MutationalPatterns standard mode had lower precision. · MAIN; Benchmarking of bioinformatics tools for…; Figure legends
ClaimedThe tool assigns SBS, doublet-base, indel and copy-number signatures (COSMIC, de novo or custom) and gives per-mutation signature probabilities. The authors state it is the first tool to offer copy-number assignment and probabilistic attribution of signatures to single mutations.
Inputs: VCF, MAF, text files, mutational matrices, and multi-sample segmentation files from ASCAT, ABSOLUTE, Sequenza, FACETS, Battenberg or PURPLE. Supports human, mouse and rat (GRCh37/38, mm9/10, rn6) and whole-genome, whole-exome and targeted sequencing. Reports cosine similarity, KL divergence, Pearson correlation and L1/L2 relative error. · MAIN; Distribution and Usage; Figure legends
Methods, data and reproducibility
Refitting is solved by nonnegative least squares (Lawson-Hanson) inside a custom forward stagewise scheme. Signatures are removed (error-increase threshold 0.01) and added back (error-decrease threshold 0.05) until nothing changes. Per-mutation probabilities come from each signature's profile weighted by its activity and divided by the reconstructed channel count.
- Previously published synthetic SBS catalogs simulating nine cancer types (bladder, esophageal, breast, lung squamous, renal, ovarian, osteosarcoma, cervical, stomach), built from 21 COSMIC signatures · synthetic; modeled on human cancer types · n = 2,700 simulated genomes (300 tumors from nine cancer types) · doi:10.6084/m9.figshare.20409430
- COSMICv3.3 SBS reference signature set, used as the full candidate input for refitting in the benchmark · n = 79 SBS signatures
- Methods & tools
- SigProfilerAssignment v0.0.28, SigProfilerMatrixGenerator, COSMIC Mutational Signatures (COSMICv3.3 SBS, DBS, ID, CN), Nonnegative least squares (Lawson-Hanson) with forward stagewise selection, deconstructSigs v1.8.0, MutationalPatterns v3.0.1 (fit_to_signatures, fit_to_signatures_strict), sigLASSO v1.1, SignatureToolsLib v2.1.2, Copy-number segmentation inputs: ASCAT, ABSOLUTE, Sequenza, FACETS, Battenberg, PURPLE
- Code & protocols
- https://github.com/AlexandrovLab/SigProfilerAssignmenthttps://github.com/AlexandrovLab/SigProfilerAssignmentRhttps://pypi.org/project/SigProfilerAssignment/https://cancer.sanger.ac.uk/signatures/assignment/https://osf.io/mz79v/wiki/home/?view_only=5998aee28bd14ef6bda9189184a984bd
- Data & accessions
- https://doi.org/10.6084/m9.figshare.20409430
- Software & environment
- Python package with an R wrapper, BSD 2-clause license, for Windows, macOS and Linux. The benchmark used SigProfilerAssignment v0.0.28, deconstructSigs v1.8.0, MutationalPatterns v3.0.1, sigLASSO v1.1 and SignatureToolsLib v2.1.2, mostly with default parameters.
- Assay conditions & compute
- SigProfilerAssignment processed 2,700 samples in 9.6 minutes (0.21 seconds per sample); CPU elapsed time and peak memory were monitored, but the benchmark hardware is not described.
- Reproducibility gaps
- The online version requires free COSMIC registration for EU/UK privacy compliance. Per-mutation probabilities are not produced from mutational vector or matrix inputs. The benchmark hardware is not reported.
Limitations (4)
- Refitting can only quantify signatures in the supplied reference set: mutational processes missing from that set cannot be found or measured, which requires de novo extraction on large cohorts. · stated by the authors
- Per-mutation signature probabilities need mutation-level input such as a VCF. If users supply a mutational vector or matrix, this output is not produced because information on individual mutations is missing. · stated by the authors
- The benchmark covered only SBS signatures on simulated genomes. Doublet, indel and copy-number assignment, per-mutation probabilities and accuracy on real tumors were not evaluated quantitatively. · noted in this summary
- A signature counted as detected if any mutation was assigned to it, so precision, sensitivity and F1 measure presence calls only. Errors in the estimated activity sizes are not reported, and the only numeric accuracy figure given is the F1 >0.90 threshold. · noted in this summary
Who should read:Read this if you need to assign COSMIC or custom SBS, doublet, indel or copy-number signatures to individual tumors or mutations, or want a head-to-head comparison of signature refitting tools.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- CANCERSIGN: a user-friendly and robust tool for identification and classification of mutational signatures and patterns in cancer genomes
Masroor Bayati, Hamid Reza Rabiee, … Hamid Alinejad‐Rokny·Scientific ReportsPeer-reviewed·Jan 28, 2020
- Uncovering novel mutational signatures by de novo extraction with SigProfilerExtractor
S. M. Ashiqul Islam, Marcos Díaz‐Gay, … Ludmil B. Alexandrov·Cell GenomicsPeer-reviewed·Sep 23, 2022
- Accurate and sensitive mutational signature analysis with MuSiCal
Hu Jin, D. Gulhan, … Peter J. Park·Nature GeneticsPeer-reviewed·Feb 15, 2024
- Computational tools to detect signatures of mutational processes in DNA from tumours: a review and empirical comparison of performance
Hanane Omichessan, Gianluca Severi, Vittorio Perduca·PLoS ONEPeer-reviewed·Sep 12, 2019
- MutSignatures: An R Package for Extraction and Analysis of Cancer Mutational Signatures
Damiano Fantini, Vania Vidimar, … Joshua J. Meeks·Scientific ReportsPeer-reviewed·Oct 26, 2020
- pyCancerSig: subclassifying human cancer with comprehensive single nucleotide, structural and microsatellite mutational signature deconstruction from whole genome sequencing
Jessada Thutkawkorapin, Jesper Eisfeldt, … Daniel Nilsson·BMC BioinformaticsPeer-reviewed·Apr 3, 2020