| name | seqeron-oncology |
| description | Cancer-genomics analysis with the Seqeron.Genomics C# API — tumor purity / ploidy (ASCAT-like), allele-specific copy number, mutational signatures (SBS96 + NMF/NNLS fitting), gene fusions, tumor mutational burden (TMB), microsatellite instability (MSI), neoantigen candidate peptides + MHC-binding classification, clonal / subclonal structure and tumor phylogeny, HRD / LOH, ctDNA / MRD. C#-API ONLY — there is NO MCP oncology server. Triggers: "estimate tumor purity / ploidy", "fit purity and ploidy (ASCAT)", "allele-specific copy number", "extract mutational signatures", "SBS96 catalog", "fit signature exposures", "call gene fusions", "is this fusion in-frame", "compute TMB", "TMB status", "MSI status / score", "predict neoantigens", "MHC-peptide binding", "classify clonal vs subclonal", "cancer cell fraction / CCF", "reconstruct tumor phylogeny", "HRD score", "LOH", "ctDNA fraction / tumor fraction", "MRD / minimal residual disease". (Germline SNP / indel calling → bio-annotation; germline read-evidence SV / CNV → seqeron-structural-variants; chromosome-arm-scale amplification / deletion / aneuploidy → bio-chromosome; oncology owns the tumor allele-specific / clonal layer.) All results are RESEARCH-GRADE, ALPHA — NOT for clinical or diagnostic use. |
| allowed-tools | Read, Bash, Grep, Glob |
seqeron-oncology — cancer genomics via the Seqeron.Genomics C# API
Routing + orchestration skill for the Oncology algorithm family (~37 documented contracts,
docs/algorithms/Oncology/*). C#-API ONLY: there is no MCP oncology server, so every pipeline
below is a real Seqeron.Genomics call chain (verified class.Method + file:line), not an MCP
tool name. This is the oncology-specific extension of seqeron-dev — it
delegates all C# mechanics (namespaces, TryCreate, the 3-tier LimitationPolicy,
SeqeronLimitationException, the Permissive test bootstrap) there and does not restate them.
- CLINICAL CAVEAT IS LOAD-BEARING HERE. Purity, HLA/MHC typing, TMB/MSI, driver/pathogenicity,
fusion, and neoantigen outputs read as clinical. They are alpha / research-grade and NOT for
clinical or diagnostic use — surface this on every reportable result (see
bio-rigor rule 6).
- Rigor is delegated to
bio-rigor: tool-only computation, provenance,
envelope, cross-check, units/coords. Do not restate — it applies by default.
- Point, don't duplicate. Contracts / invariants live in
docs/algorithms/Oncology/*.md; the
operating envelope in docs/Validation/LIMITATIONS.md. Link, never copy.
Namespace & shape (verified in source)
Everything below lives in one namespace Seqeron.Genomics.Oncology, across three static classes:
| Class | Covers | Source (from repo root ../../../) |
|---|
OncologyAnalyzer | purity/ploidy/ASCAT, copy number, signatures, fusions, TMB/MSI, neoantigen windows, matrix pMHC, clonal/phylogeny, HRD/LOH, ctDNA | src/Seqeron/Algorithms/Seqeron.Genomics.Oncology/OncologyAnalyzer.cs:30 |
ImmuneAnalyzer | immune-infiltration / deconvolution + ESTIMATE-purity (guarded ONCO-IMMUNE-001) | src/.../Seqeron.Genomics.Oncology/ImmuneAnalyzer.cs:23 |
MhcflurryAffinityPredictor | ANN IC50 prediction (caller supplies the weight pack) | src/.../Seqeron.Genomics.Oncology/MhcflurryAffinityPredictor.cs:45 |
OncologyAnalyzer is a large static class; find any method with
grep -n "public static" src/Seqeron/Algorithms/Seqeron.Genomics.Oncology/OncologyAnalyzer.cs.
For the full curated Method-ID index see reference/api-conventions.md.
Guarded units — STOP rule (in scope)
Both throw SeqeronLimitationException when the effective mode is below Moderate (i.e. under
Strict). Default mode is Moderate, so they run by default; a Strict process must STOP and
report (name the unit + workaround), never widen to force output. Read
bio-rigor envelope + docs/Validation/LIMITATIONS.md.
- ONCO-MHC-001 — matrix (SMM/BIMAS) pMHC scorers
PredictIc50Smm / PredictBindingHalfLifeBimas
each call LimitationPolicy.Enforce("ONCO-MHC-001") (OncologyAnalyzer.cs:8523, :8490).
No redistributable matrix ships — the caller supplies the coefficients. Workaround: use a
vendor model (NetMHCpan-4.1) or ClassifyMhcBinding on a caller-supplied IC50.
- ONCO-IMMUNE-001 —
ImmuneAnalyzer.EstimateTumorPurity / DeconvoluteImmuneCells[NuSvr]
call LimitationPolicy.Enforce("ONCO-IMMUNE-001") (ImmuneAnalyzer.cs:477, :525, :680).
Ships a bundled ABIS matrix, not exact CIBERSORT-LM22; ESTIMATE purity is Affymetrix-calibrated.
- ONCO-ASCAT-001 / ONCO-PURITY-001 are documented-only (no runtime guard) — the purity/ploidy
grid fit is exact for its stated ASCAT-style model; report the model caveat, nothing throws.
Decision guide — task → C# entry point
| Task | Method ID (OncologyAnalyzer.* unless noted) · file:line |
|---|
| Fit tumor purity + ploidy jointly (ASCAT grid) | FitPurityPloidy · :7411 |
| Quick purity from VAF cluster | EstimatePurity · :515 / EstimatePurityFromVAF · :451 |
| Ploidy from allele-specific segments | EstimatePloidy · :6554 |
| Segment allele-specific CN (ASPCF) | SegmentAlleleSpecificAspcf · :7641 / SegmentAlleleSpecific · :7282 |
| log2-ratio → copy number; classify amp/del | Log2RatioToCopyNumber · :5900 → ClassifyCopyNumbers · :5995 |
| SBS96 channel of a mutation / all 96 channels | ClassifySbsContext · :2659 / EnumerateSbs96Channels · :2696 |
| Extract signatures de novo (NMF) | ExtractSignatures · :3431 |
| Fit / refit known signatures (NNLS) | FitSignatures · :2933 → match: MatchToReferenceSignatures · :4609 |
| Bootstrap exposure confidence intervals | BootstrapExposures · :4774 |
| Call fusions; is it in-frame? | DetectFusions · :5421 / IsInFrame · :5382 |
| Known-fusion DB lookup; breakpoint analysis | MatchKnownFusions · :5558 / AnalyzeBreakpoint · :5703 |
| TMB (count or calls) + status | CalculateTMB · :1508/:1540 → ClassifyTMB · :1557 |
| MSI score + status (Bethesda) | CalculateMSIScore · :1634 → ClassifyMSIStatus · :1664 / ClassifyBethesdaPanel · :1688 |
| Neoantigen candidate peptide windows | GenerateNeoantigenPeptides · :8026 |
| Classify a supplied IC50 → binding strength | ClassifyMhcBinding · :8292 |
| ANN IC50 (caller weight pack) | MhcflurryAffinityPredictor.PredictIc50 · MhcflurryAffinityPredictor.cs:619 |
| Clonal vs subclonal; CCF | ClassifyClonality · :6789 / EstimateCcf · :7014 |
| CCF clustering → subclones; heterogeneity | ClusterCcfValues · :7061 / AnalyzeHeterogeneity · :11042 |
| Reconstruct tumor phylogeny | ReconstructPhylogeny · :10617 |
| HRD score / LOH | CalculateHrdLohScore · :2017 / DetectLOH · :1979 |
Canonical C#-API pipelines
(a) Tumor purity + ploidy from allele-specific segments (ASCAT-like)
SegmentAlleleSpecificAspcf(...) → AlleleSpecificSegmentSummary[] (OncologyAnalyzer.cs:7641).FitPurityPloidy(segments, purityMin=0.05, purityMax=1.0, purityStep=0.01, ploidyMin=1.5, ploidyMax=5.0, ploidyStep=0.05, gamma=…) → PurityPloidyFit (:7411).- Cross-check purity via an independent VAF path:
EstimatePurity(variants) (:515) — magnitudes should agree.
Provenance
1) OncologyAnalyzer.SegmentAlleleSpecificAspcf(logR,baf,positions,…) → segments
2) OncologyAnalyzer.FitPurityPloidy(segments, grid defaults) → {purity, ploidy, gof}
3) OncologyAnalyzer.EstimatePurity(vafVariants) → purity (independent cross-check)
Envelope: ONCO-ASCAT-001 / ONCO-PURITY-001 documented-only — exact for the ASCAT-style grid model.
Caveat: alpha, research-grade — NOT for clinical use; purity is decision-relevant, validate independently.
(b) Mutational signatures — SBS96 catalog → fit → confidence
- Per mutation:
ClassifySbsContext(fivePrime, ref, alt, threePrime) → one of 96 channels (:2659); build the 96-vector catalog (EnumerateSbs96Channels :2696 gives the canonical order).
FitSignatures(catalog, referenceSignatures) → NNLS exposures (:2933); or ExtractSignatures(...) for de-novo NMF (:3431).MatchToReferenceSignatures(...) (:4609) to label; BootstrapExposures(...) (:4774) for CIs — report exposures with their intervals.
(c) Gene fusions — call → frame → annotate
DetectFusions(candidates, thresholds?) → FusionCall[] (:5421).- Per call:
IsInFrame(fivePrimeCodingBases, threePrimeStartPhase) (:5382); MatchKnownFusions(...) (:5558) for DB hits; AnalyzeBreakpoint(...) (:5703) for junction detail.
(d) TMB + MSI status
CalculateTMB(mutationCount, targetRegionMb) (:1508) → ClassifyTMB(tmb) → TmbStatus (:1557).CalculateMSIScore(unstableLoci, totalLoci) (:1634) → ClassifyMSIStatus(score) (:1664) (or ClassifyBethesdaPanel :1688).
- Both are clinical-flavoured — attach the caveat; report units (TMB = mutations/Mb).
(e) Neoantigen candidates + MHC binding (guarded scorer)
GenerateNeoantigenPeptides(wildTypeProtein, mutantResidue, mutationPosition, minLength=8, maxLength=11) → mutant peptide windows (:8026).- Score binding: preferred ANN
MhcflurryAffinityPredictor.PredictIc50(networks, peptide, allele) (MhcflurryAffinityPredictor.cs:619) with a caller weight pack; then ClassifyMhcBinding(peptideLength, ic50Nm, mhcClass) (:8292).
- Matrix scorers
PredictIc50Smm (:8516) / PredictBindingHalfLifeBimas (:8483) are guarded (ONCO-MHC-001) and need caller-supplied coefficients — if unavailable, STOP and report.
End-to-end grounded example
Task. From allele-specific segments + somatic VAFs, estimate purity/ploidy, then place mutations on the clonal timeline.
SegmentAlleleSpecificAspcf(...) → segments (:7641).FitPurityPloidy(segments, defaults) → {purity, ploidy} (:7411).ClassifyClonality(clonalityVariants, purity) → clonal vs subclonal (:6789); per variant EstimateCcf(vaf, purity, tumorCN, multiplicity) (:7014).ClusterCcfValues(ccfValues, clusterCount) (:7061) → ReconstructPhylogeny(...) (:10617) for the subclone tree.
Provenance
1) SegmentAlleleSpecificAspcf(logR,baf,pos) → segments
2) FitPurityPloidy(segments) → purity,ploidy
3) ClassifyClonality(variants,purity); EstimateCcf(vaf,purity,cn,mult) per variant
4) ClusterCcfValues(ccf,k) → ReconstructPhylogeny(...) → subclone tree
Envelope: ONCO-ASCAT-001/ONCO-PURITY-001 documented-only (ASCAT-style model, exact for that model).
Caveat: alpha, research-grade — NOT for clinical/diagnostic use; validate independently.
Reference
