bio-clinical-databases-somatic-signatures

What it does

Extracts and assigns COSMIC v3.4 mutational signatures (84 SBS / 11 DBS / 18 ID / 24 CN / 16 SV) from somatic VCFs and maps signatures to etiology, with options for de novo extraction or refit to COSMIC depending on cohort size. Supports BRCA1/2 HRD, MMR-D, POLE, APOBEC3A, UV, tobacco, aflatoxin, 5-FU/SBS17b, platinum, colibactin SBS88 etiology and informs PARP inhibitor decisions. Uses a mix of tools including SigProfilerSuite, MutationalPatterns, MuSiCal mvNMF, SigNet, or HRDetect. Provides guidance on when to use de novo vs refit based on mutation counts and cohort size, and includes a workflow that pins versions and performs stability checks for SigProfilerExtractor."

How it works

• The skill outlines a three-step workflow under Standard Workflow: SigProfilerSuite: install reference genome, generate a 96-context matrix from VCFs, then perform either de novo extraction with stability gates (SigProfilerExtractor) or refit to COSMIC (SigProfilerAssignment) depending on cohort size. The refit path uses SigProfilerAssignment with signature_database alignment, NNLS penalties, and plotting outputs.
• For R users, MutationalPatterns is used for strict refit or de novo extraction via extract_signatures and fit_to_signatures.
• HRDetect is provided as a BRCA1/2 deficiency classifier using six features and a lasso-based approach.
• The description details tooling compatibility, version references, and example code blocks for both Python and R environments.
• It includes a decision-tree guide for choosing between de novo vs refit, thresholds based on mutation counts per sample and cohort size, and stability criteria for de novo extraction (nmf_replicates, stability thresholds, and cosine-based criteria).

When to use it

• Use when characterizing DNA-damage etiology (e.g., BRCA1/2 HRD, MMR-D, POLE, APOBEC3A, UV, tobacco, aflatoxin, 5-FU/SBS17b, platinum, colibactin SBS88) to inform clinical decisions or audit de novo extraction vs refit in cohorts of various sizes.
• Use HRDetect for BRCA1/2 deficiency screening and as a BRCA-deficiency classifier.
• Use de novo extraction for cohorts expected to harbor novel signatures; use refit for established cancer types or small cohorts; use low-mutation-count strategies like SigNet when counts are very low.

What it can touch

• The workflow references tool usage such as SigProfilerMatrixGenerator, SigProfilerExtractor, SigProfilerAssignment, MutationalPatterns, HRDetect, SigNet, MuSiCal mvNMF. It specifies parameters like nmf_replicates=100 and cosmic_version=3.4 for stability and alignment with COSMIC v3.4.

Caveats

• It notes compatibility constraints and version notes for several tools and emphasizes stability gates for SigProfilerExtractor (nmf_replicates=100, minimum stability requirements).
• It cites potential limitations such as heavy compute for de novo extraction and strict refit behavior for small cohorts or single-sample analyses.
• It addresses FFPE artifact considerations and the need for explicit FFPE handling in certain scenarios.
• It mentions the deprecation of deconstructSigs and recommends SigProfilerAssignment or MutationalPatterns for strict refits.