Prostate cancer options catalog

MethylPipeline is disease-agnostic. Prostate cancer is the primary worked example—not a separate PRAD pipeline. This chapter lists what researchers can choose (analyte, assay, alignment, extraction, cohort design, statistical axis, SaMD tier) and where each choice is set.

For why those choices matter to physicians and payers, see Prostate Cancer Application Deep-Dive. The generic application-pack pattern is ch.24. Alzheimer (ch.21) documents one procedure; this chapter is a multi-path catalog.

Claim boundary. Architecture, profiles, and filled feasibility packages are not FDA clearance, CLIA validation, or coverage-ready clinical performance. Registered packages EV-PCA-PLASMA-2026-06, EV-PCA-HGOOD-2026-06, and EV-PCA-BUFFY-2026-07 are feasibility / engineering evidence. They lack patient-disjoint locked_test / pivotal_validation partitions and must not be cited for gatekeeper NPV, biopsy deferral, or payer medical-necessity claims. See validation-evidence-index and SaMD lifecycle (ch.18).

flowchart TB
  goal[ResearcherGoal] --> analyte{AnalyteAndAssay}
  analyte -->|host_response| buffy[buffy_wgbs_pangenome_gene_fc]
  analyte -->|plasma_discovery| plasma[cfdna_wgbs_plasma]
  analyte -->|paired_liquid| both[TwoProceduresMatchedIDs]
  analyte -->|gatekeeper| emseq[cfdna_emseq_targeted]
  analyte -->|tumor_progression| tissue[tissue_analyte_linear_WGBS]
  buffy --> design[BinaryOrStagedGleason]
  plasma --> design
  both --> design
  emseq --> design
  tissue --> design
  design --> axis[researchMode]
  axis --> ladder[samd_research_to_pivotal]

Prostate research and diagnostic option tree

1. Start here — pick a research or diagnostic goal

Intent	Analyte	Procedure / run recipe	Typical profile	Do not claim from this alone
Host / immune / aggression research	buffy_coat	buffy_wgbs_pangenome_gene_fc (linear / Mojo / residual alternates below)	samd_research + researchMode: gene_fc	Buffy as tumor-shed NPV / biopsy deferral
Genome-wide plasma discovery	cfdna	cfdna_wgbs_plasma	samd_research + gene_fc	Discovery BA as a gatekeeper operating point
Paired buffy + plasma	both	Two studies, matched patient_id	Same ladder per study	Unpaired buffy-trained → plasma-deployed without retrain
Pre-biopsy GG≥2 geometry	cfdna	cfdna_emseq_targeted + operator target_panel_bed	Climb to samd_pivotal with partitions	Feasibility WGBS healthy-vs-pooled-PCa BA
Tumor / adjacent-normal for progression and panel anchoring	tissue	No shipped procedure — primary_analyte: tissue + linear WGBS SamplePrep (see §5b)	samd_research; optional cell_deconv_hitimed	Tissue-only as a liquid-biopsy intended use

Set regulatory.primary_analyte on the study manifest so it matches the procedure analyteExpectation. Do not mix tumor DNA into a buffy or cfDNA primary-analyte cohort. Tool knobs stay on site / procedure / profile / instance overlay — never in project_*.json (step_config is rejected).

Merge order (highest wins): instance → program with / stepOverride → procedure → profile / mode → analyte fill-missing → site. See config-parameter-matrix.

2. Assay procedure options

Shipped packs live under workflow_engine/domain/profiles/procedures/. All default to pipelineProfile: samd_research and researchMode: gene_fc.

Procedure	Analyte	Align / protocol	Lifecycle	When to use	When not
buffy_wgbs_pangenome_gene_fc	buffy_coat	pangenome_wgbs / wgbs_pangenome	study_validation_lifecycle (Houseman)	Default buffy ~30× WGBS research	Tumor-shed rule-out; missing HPRC d9-bs assets
buffy_wgbs_linear_gene_fc	buffy_coat	linear (Clara parabricks) / wgbs_linear	same	Faster / cheaper linear baseline; NVIDIA Clara path	When pangenome_wgbs is already accepted and graph assets are up
buffy_wgbs_linear_mojo_gene_fc	buffy_coat	linear (Mojo) / wgbs_linear	same	Portable NVIDIA / AMD / CPU linear	Clara-only sites that want the explicit Parabricks image
buffy_wgbs_mvalue_residual_gene_fc	buffy_coat	pangenome_wgbs	study_validation_lifecycle_residual	Confounder-adjusted host DMPs (Ω ALR + smoking / age / BMI / CRP)	Default buffy path; residualization is opt-in
cfdna_wgbs_plasma	cfdna	linear / wgbs_linear	study_validation_lifecycle_no_deconv	Plasma discovery + fragmentomics	Buffy host biology; EM-Seq gatekeeper geometry
cfdna_emseq_targeted	cfdna	linear / emseq_targeted	no_deconv + sample_prep_emseq	Locked panel, elevated min_cov, no genome-wide DMP hunt	Genome-wide discovery; missing target_panel_bed
cfdna_emseq_mhl_survival	cfdna	linear / emseq_targeted	study_validation_mhl_survival + sample_prep_emseq	Wong-style MHB/MHL + Cox OS (operator target_panel_bed + survival_path)	FeatureCuts/ECDF classification; missing haplotype sidecars

Tumor tissue — no shipped procedure. The catalog analyte tissue.analyte.json exists (family: analyte, visibility advanced). Set regulatory.primary_analyte: tissue and reuse sample_prep.program.json + study_validation_lifecycle.program.json (or the no-deconv fork if you skip HiTIMED). Prefer linear WGBS for FFPE / fresh-frozen tumor BAMs. A future tumor_wgbs_linear_gene_fc procedure is a follow-on — do not invent that ID in run context today.

Linear vs pangenome_wgbs is decided by the compare harness (sample_prep_test_bed, scripts/compare_sample_prep_linear_vs_wgbs.sh), not by classifier BA alone. Until a report shows both arms through extraction QC, keep both buffy procedures available. Stock pangenome / Giraffe is an engineering comparator only — no WGBS methylation parity.

3. Alignment options

Details: alignment-engines, Sample prep and QC (ch.03).

Choice	Values	Where set	PCa note
alignmentMode	linear, pangenome_wgbs, stock pangenome	Procedure (alignmentMode)	Plasma and EM-Seq default linear; buffy research default pangenome_wgbs
Linear engine	parabricks (Clara) or mojo	actionConfig.parabricks.engine	buffy_wgbs_linear_gene_fc vs buffy_wgbs_linear_mojo_gene_fc
WGBS graph engine	gpu_giraffe / mojo_giraffe, cpu_vg	actionConfig.methylgrapher_wgbs.align_engine	Production GPU path vs vg giraffe rollback
GPU fallback	mojo | vg | error	gpu_giraffe_fallback	Unknown vendor or explicit fail-closed
Linear reference	GRCh38 FASTA	Site reference_genome / linear_genome_key	Human PCa studies
WGBS pangenome	HPRC d9-bs C2T+G2A	Site methylgrapher_wgbs / pangenome_wgbs_key	Required for buffy pangenome procedure
Library protocol	wgbs_linear, wgbs_pangenome, emseq_targeted	Procedure libraryProtocol	EM-Seq is assay design, not an aligner bake-off

Site-published core guardrails (alignment_qc.core_guardrails) gate eligibility: PF / Q30 / mean quality, AT/GC dropout, insert-size window, deamination / OxoG. Sparse overlays belong on profile / procedure / study via sample_prep_guardrails_overlay. cfDNA adds fragmentomics QC when the analyte profile fills it. EM-Seq emphasizes on-target depth against the locked BED.

4. Extraction options

Backend follows alignment: sample.methyl_extract (MethylExtractor) for linear / stock pangenome; sample.methylgrapher_wgbs_extract for pangenome_wgbs. Schema: methyl_extract.schema.json. CLI mirror: MethylExtractor README.

Option	Typical PCa setting	Where set
Contexts	CG always; CHG/CHH optional for discovery	Study contexts + profile extract_contexts
Min coverage	WGBS site/profile default; EM-Seq min_cov: 20 in the procedure	Procedure / site actionConfig.methyl_extract
MAPQ / Phred	Site or profile (-q / -p)	Same
Cap coverage	Off unless PCR-bias correction is wanted	Same
Read-level patterns	On for WGBS informME (tile_size 4 in buffy/plasma procedures); off for EM-Seq	Procedure methyl_extract.read_level
Target panel BED	Required for EM-Seq	Study / application overlay target_panel_bed — not Python
Output	Per-chr {ctx}.h5 (+ optional .patterns.h5, TSV)	output_format

Extraction QC (extraction_qc.guardrails) is site-published: CpG weighted mean coverage, mammalian CHG/CHH caps, autosomal uniformity, discard fraction. The extractor sidecar contract lives in the sibling MethylExtractor repo (docs/extraction_qc_contract.md).

5. Study-design options

Production manifests live under /work/projects/prostate-cancer/configs/. Repo project_*.json files under workflow_engine/domain/checks/ and tools/methyl-config-editor/configs/ are CI / editor smoke mirrors, not the production source of truth.

Design	Example tree	Comparisons	Notes
Binary buffy	Buffy_healthy_vs_PCa, H_PCa_good	Healthy vs PCa	CI: checks/buffy_healthy_vs_pca, checks/h_pca_good
Binary plasma	Plasma_healthy_vs_PCa	Healthy vs PCa	cfDNA analyte + fragmentomics
Staged Gleason	Healthy_vs_PCa1-5-CG	control_vs_each_disease or explicit pairs	CI: checks/pca1_5_cg; profile staged_ovr_mc when running multi-class OvR
Progression	any staged manifest	progression_order + progression_labels	runProgressionAnalysis: true

Prostate overlay knobs (instance / context_*.json, not the study manifest science block):

Key	Typical PCa value
pipelineProcedure	One of the shipped IDs in §2
actionConfig.mapper.disease_term	"Prostate Cancer"
actionConfig.mapper.enrich_disease	true for human disease priors
actionConfig.enricher.library_preset	cancer-core (cfDNA analyte default) or cancer-extended (staged profiles)
actionConfig.progression.enabled / runProgressionAnalysis	true on staged studies
actionConfig.progression.disease_context	prostate_cancer — bundled AR / DNA-repair / replication / metabolic gene sets in profiles/prostate_cancer.json

Scaffold:

source .venv/bin/activate
methyl-study-init \
  --study-id prostate-cancer \
  --name Buffy_healthy_vs_PCa \
  --analyte buffy_coat \
  --binary \
  --output-root /work/projects
# or: --analyte cfdna|tissue   --stages 5

--analyte for methylation is cfdna | buffy_coat | tissue | plant_tissue. Fill CSVs with sample IDs; assign validation_partitions early (ch.18).

5b. Tumor tissue when analyzing disease progression

Progression scoring is comparison/stage-driven (mapper combined CSVs per Gleason / stage label), not analyte-aware. Tumor DNA is the natural matrix for grade-ordered progression; liquid and buffy remain complementary host / shed layers.

Design	Manifest shape	Why for PCa progression
Tissue-only staged	primary_analyte: tissue; diseases.stages[] = adjacent-normal (optional control) + GG1…GG≥2 / PCa1–5 from RP or targeted biopsy	Direct tumor epigenome vs grade; best input to disease_context: prostate_cancer
Adjacent-normal vs tumor	Binary or two-stage tissue study (benign / uninvolved vs index lesion)	Field-effect vs tumor-specific DMPs. Historical packages/methylcentroid/configs/tissue_benign_config.json / tissue_malignant_config.json are legacy package configs, not production manifests
Tissue-anchored then liquid transfer	Discover / freeze on tissue; lock target_panel_bed or FeatureCuts panel; retrain on matched plasma (optional buffy) with enforce_training_analyte_match	GSTP1-style tissue markers → cfDNA / EM-Seq gatekeeper. Do not deploy a tissue-trained classifier on plasma without retrain
Matched trio	Three studies, same patient_id / independence_keys: tissue + buffy_coat + cfdna	Tumor = grade ground truth and purity; buffy = hematopoietic / CHIP analogue; plasma = shed signal. combined is not a substitute — it only fills bisulfite defaults and mixes QC poorly
Tissue + HiTIMED purity	Overlay cell_deconv_hitimed (method: hitimed, analyte: tissue) + operator hierarchy_basis_path	Tumor / stroma as ECDF covariates so progression is not purity-confounded. Blood immune subtree ships in-wheel; prostate tumor/stromal atlas does not — provision offline (ANALYTE_PROFILES, HiTIMED plan)

Honesty:

• normalize_primary_analyte("tissue") passes through; analyte_profiles.py has no tissue branch — fill-missing is bisulfite-only (same as combined). Set fragmentomics off, mammalian CHG/CHH caps, and HiTIMED on the profile / instance overlay.
• FFPE vs fresh-frozen is a preanalytics / QC overlay (insert size, deamination, coverage uniformity), not a new analyte. Do not invent an ffpe token.
• Tissue progression does not make EV-PCA packages diagnostic. A tissue-anchored panel still needs liquid-biopsy holdout / pivotal partitions for a pre-biopsy claim.
• Do not put tumor and buffy samples in one primary_analyte cohort.

6. Statistical axis and modeling options

researchMode	DMP axis	Gene axis	Typical PCa use
dmp_raw	raw_pool	none	DMP recurrence only
dmp_fc	featurecuts	none	BA-gated DMP panel
gene_enricher	raw_pool	mapper / enricher recurrence	Pathway gene stability
gene_fc	raw_pool	featurecuts	Default on all shipped PCa procedures
dual_fc	featurecuts	featurecuts	Full DMP + gene panels

Modes live under workflow_engine/domain/profiles/modes/. Deprecated mc_* profile names fold into samd_research + a mode — prefer the explicit pair.

Deconvolution

Method	Analyte tree	PCa default
Houseman	6-cell Ω (buffy)	Buffy procedures
HiTIMED	Buffy immune subtree; cfDNA tumor_fraction leaf; tissue full tumor/immune/stromal	Overlay cell_deconv_hitimed; cfDNA/tissue tumor leaves need operator atlases
None	—	Plasma WGBS and EM-Seq lifecycles

Also choose (site / profile, not Python defaults): FeatureCuts BA targets and gene-selection caps (gene_selection.max_dmps / max_genes); informME and derived_measures as ECDF second-stage covariates; model backends ECDF (preferred regulated path), tabular_sklearn, optional hybrids. Hyperparameter search (ch.15) runs on a fixed procedure — do not grid pipelineProcedure.

Stability / freeze / model: ch.05–ch.08.

7. Research vs diagnostic (SaMD ladder)

Intent	Profile	Required partitions	Allowed narrative
Feasibility / method	samd_research	locked_test recommended	Engineering, panel stability, method development
Internal validation	samd_holdout_enrichment	non-empty locked_test	Locked-HP enrichment
Claim-oriented	samd_pivotal	non-empty pivotal_validation	Clinical performance only when evidence is filled

Clinical-performance claims (allow_clinical_performance_claims: true) are code-blocked before pivotal_validation. SOP: ch.18.

Current EV-PCA status (validation-evidence-index):

Package	Study tree	Analyte	Status
EV-PCA-PLASMA-2026-06	Plasma_healthy_vs_PCa	cfDNA	Draft feasibility
EV-PCA-HGOOD-2026-06	H_PCa_good	buffy	Draft feasibility; gene-stable-at-threshold gaps reported
EV-PCA-BUFFY-2026-07	Buffy_healthy_vs_PCa	buffy	Draft; incomplete model chain for gatekeeper claims

The diagnostic path for a pre-biopsy claim is EM-Seq (or proven plasma WGBS) + partitions + pivotal — not buffy-only NPV, and not tissue-only. Tissue informs panel and grade biology; the gatekeeper intended use remains a liquid biopsy.

8. Worked commands

Activate the repo venv first. Prefer samd_research + pipelineProcedure over deprecated mc_* profiles. Use staged_ovr_mc only for Gleason multi-class OvR.

Buffy pangenome research (binary)

source .venv/bin/activate
methyl-workflow-run \
  --program workflow_engine/domain/fixtures/study_validation_lifecycle.program.json \
  --context '{
    "projectPath":"/work/projects/prostate-cancer/configs/project_Buffy_healthy_vs_PCa.json",
    "pipelineProcedure":"buffy_wgbs_pangenome_gene_fc",
    "pipelineProfile":"samd_research",
    "researchMode":"gene_fc"
  }' \
  --parallel-workers 1

Plasma WGBS discovery

methyl-workflow-run \
  --program workflow_engine/domain/fixtures/study_validation_lifecycle_no_deconv.program.json \
  --context '{
    "projectPath":"/work/projects/prostate-cancer/configs/project_Plasma_healthy_vs_PCa.json",
    "pipelineProcedure":"cfdna_wgbs_plasma",
    "pipelineProfile":"samd_research",
    "researchMode":"gene_fc"
  }' \
  --parallel-workers 1

Staged Gleason OvR (buffy)

methyl-workflow-run \
  --program workflow_engine/domain/fixtures/mc_stability_staged.program.json \
  --context '{
    "projectPath":"/work/projects/prostate-cancer/configs/project_Healthy_vs_PCa1-5-CG.json",
    "pipelineProfile":"staged_ovr_mc"
  }' \
  --parallel-workers 1

EM-Seq gatekeeper geometry (set target_panel_bed on the overlay first)

methyl-workflow-run \
  --program workflow_engine/domain/fixtures/study_validation_lifecycle_no_deconv.program.json \
  --context '{
    "projectPath":"/work/projects/prostate-cancer/configs/project_Plasma_healthy_vs_PCa.json",
    "pipelineProcedure":"cfdna_emseq_targeted",
    "pipelineProfile":"samd_research",
    "researchMode":"gene_fc"
  }' \
  --parallel-workers 1

Tissue staged progression (no shipped procedure — analyte + programs)

methyl-study-init \
  --study-id pca-tissue \
  --name Adjacent_vs_PCa_stages \
  --analyte tissue \
  --stages 5 \
  --output-root /work/projects

methyl-workflow-run \
  --program workflow_engine/domain/fixtures/study_validation_lifecycle.program.json \
  --context '{
    "projectPath":"/work/projects/pca-tissue/configs/project_Adjacent_vs_PCa_stages.json",
    "pipelineProfile":"samd_research",
    "researchMode":"gene_fc",
    "runProgressionAnalysis": true,
    "actionConfig": {
      "progression": {"enabled": true, "disease_context": "prostate_cancer"},
      "mapper": {"disease_term": "Prostate Cancer", "enrich_disease": true},
      "enricher": {"library_preset": "cancer-extended"},
      "fragmentomics": {"enabled": false}
    }
  }' \
  --parallel-workers 1

Add a cell_deconv_hitimed overlay (method: hitimed, analyte: tissue, operator hierarchy_basis_path) when purity covariates are required.

Validate before enrichment or pivotal:

methyl-study-validate-manifest \
  --project /work/projects/prostate-cancer/configs/project_Buffy_healthy_vs_PCa.json \
  --profile samd_research

9. Existing fixtures and /work tree

Location	Role
/work/projects/prostate-cancer/	Production study tree (manifests, cohort CSVs, MC outputs)
/work/projects/prostate-cancer/configs/project_*.json	Operator-edited study manifests
workflow_engine/domain/checks/buffy_healthy_vs_pca/	CI smoke: binary buffy
workflow_engine/domain/checks/pca1_5_cg/	CI smoke: staged PCa1–5
workflow_engine/domain/checks/h_pca_good/	CI smoke: H_PCa_good
workflow_engine/domain/fixtures/*.program.json	Topology (sample prep, MC, lifecycle)
tools/methyl-config-editor/configs/	Editor copies — not production SoT

Edit cohorts on /work. Programs and profiles stay in the repo or the promoted runtime-bundle. See work layout and config registry (ch.19).

Not shipped in this pass: docs/examples/samd/prostate-* stubs and a dedicated tumor-tissue procedure. Follow ch.24 if you add those later.

Further reading

• Physician / payer narrative: Prostate Cancer Application Deep-Dive
• Analyte biology: BuffyCoat vs cfDNA, ANALYTE_PROFILES
• Gatekeeper clinical SOW: Prostate Cancer Detection
• Fitness vs that SOW: Prostate_Cancer_Detection_MethylPipeline_Fitness
• Wong et al. 2026 EM-Seq mCRPC prognosis (MHB/MHL + OS nomogram): use procedure cfdna_emseq_mhl_survival + study survival_path. Gap note (implemented / remaining): Wong2026_EMSeq_mCRPC_MethylPipeline_Gaps. Florida community-access framing: partnership brief
• Tutorial (healthy vs stages): ch.16
• Command cookbook: ch.12