24 methylation application packs

An application pack is a study configuration on an existing omics process pack (cohorts, partitions, and a small config overlay)—not a new modality. Between process and application sits an assay procedure pack: a named recipe for how the matrix is sequenced and scored (library protocol, SamplePrep, covariates, FeatureCuts axis).

This chapter is the generic pattern. Worked instances:

Instance	Kind	Guide
Alzheimer cfDNA (Control → MCI → AD)	Disease application	ch.21
Plant abiotic stress (Arabidopsis + crop sites)	Trait application	ch.23
Prostate cancer (buffy / plasma / EM-Seq / tissue)	Disease application (multi-path catalog)	ch.25

flowchart TB
  subgraph process [Process packs]
    METH[methylation WGBS]
    RNA[rnaseq]
    PROT[proteomics]
  end
  subgraph apps [Application packs on methylation]
    AD[Alzheimer cfDNA]
    PLANT[Plant abiotic stress]
    PCA[Prostate cancer]
    FUTURE[Future indication or trait]
  end
  METH --> AD
  METH --> PLANT
  METH --> PCA
  METH --> FUTURE

Process packs vs methylation application packs

Process vs procedure vs application¶

	Process pack	Assay procedure pack	Application pack
What it adds	Omics modality: DomainPrograms, typed actions, QC	Library protocol + SamplePrep/lifecycle pointers + science defaults (aligner, informME, deconv, FeatureCuts)	Indication/trait: cohorts, partitions, disease/trait overlay, enrichment preset
Typical change	New aligner / feature contract	New `*.procedure.json` (rarely a thin program fork)	New `project_.json` + `context_.json`
Examples	Methylation WGBS, RNA-Seq, proteomics	`buffy_wgbs_pangenome_gene_fc`, `cfdna_wgbs_plasma`, `cfdna_emseq_targeted`	Alzheimer cfDNA, plant drought stress, prostate cancer
Operator surface	`regulatory.primary_modality`	`pipelineProcedure`	Same modality + procedure + overlay / analyte

Informal subtypes (disease application, trait application) are fine in prose; checklists and hub docs use application pack.

Shipped methylation assay procedures¶

Committed under workflow_engine/domain/profiles/procedures/:

Procedure	Analyte	Defaults
`buffy_wgbs_pangenome_gene_fc`	`buffy_coat`	Default WGBS pangenome (native-Mojo methylGrapher on NVIDIA or AMD, `alignmentMode: pangenome_wgbs`); requires `hprc-d9-bs.wl.gfa`
`buffy_wgbs_mvalue_residual_gene_fc`	`buffy_coat`	Opt-in M-value residualization before centroids/DMPs (Houseman Ω ALR + smoking/Hannum age/BMI/CRP scores). Forked lifecycle. The research note is buffy-mvalue-residualization.md.
`buffy_wgbs_linear_gene_fc`	`buffy_coat`	Linear WGBS via Clara Parabricks (`parabricks.engine` `parabricks`)
`buffy_wgbs_linear_mojo_gene_fc`	`buffy_coat`	Same linear program node via GoliathAlign (`parabricks.engine` `mojo`)
`cfdna_wgbs_plasma`	`cfdna`	Linear WGBS + fragmentomics (analyte merge), gene FeatureCuts, no cell deconv lifecycle
`cfdna_emseq_targeted`	`cfdna`	`libraryProtocol: emseq_targeted`, `sample_prep_emseq`, panel BED + elevated `min_cov`, no deconv
`cfdna_emseq_mhl_survival`	`cfdna`	Same SamplePrep; `researchMode: mhl_survival`; `pipeline.mhb_mhl` + Cox (not FeatureCuts/ECDF)
`plant_wgbs_gene_fc`	`plant_tissue`	Linear WGBS, gene FeatureCuts, plant lifecycle (no blood deconv)

methyl-workflow-run \
  --program workflow_engine/domain/fixtures/study_validation_lifecycle.program.json \
  --context '{
    "projectPath":"/work/projects/my-buffy/configs/project_….json",
    "pipelineProcedure":"buffy_wgbs_pangenome_gene_fc",
    "pipelineProfile":"samd_research"
  }'

Use buffy_wgbs_pangenome_gene_fc for WGBS pangenome buffy research, buffy_wgbs_linear_gene_fc for linear Clara, and buffy_wgbs_linear_mojo_gene_fc for linear GoliathAlign. The start command and the four align paths are in Start a SamplePrep instance.

Prefer the procedure’s lifecycleProgram / samplePrepProgram hints (especially study_validation_lifecycle_no_deconv for cfDNA plasma and EM-Seq). Schema: schemas/config/procedure.schema.json.

EM-Seq note. Set actionConfig.methyl_extract.target_panel_bed to an operator- supplied panel BED (cancer-specific panels belong in the application pack or study, not in Python). Extract restricts the BAM with samtools view -L before MethylExtractor.

EM-Seq MHL + survival is a parallel procedure (cfdna_emseq_mhl_survival, researchMode: mhl_survival). It keeps the same SamplePrep, emits {chrom}-CG.mhap.h5 haplotype sidecars, runs pipeline.mhb_mhl, and models overall survival with the Cox backend. It does not replace cfdna_emseq_targeted + gene FeatureCuts. Study manifests need survival_path (CSV: sample_id, time, event, optional labs). Haplotype contract: mhap_store_contract.md.

Required artifacts (application pack)¶

Artifact	Role
Study manifest (`project_*.json`)	Groups/comparisons (or stages), chromosomes, contexts, `regulatory`, `validation_partitions`
Context overlay (`context_*.json`)	`pipelineProcedure` + disease/trait `actionConfig` + `pipelineProfile`
Cohort CSVs	Sample ID lists under `data/` (plant- or patient-disjoint as appropriate)
Enrichment preset (optional)	Named entry in `library_presets.json` when human `cancer-core` is wrong
Analyte / site (as needed)	e.g. `cfdna` vs `plant_tissue`; non-human genomes via site pins
Lifecycle program	Taken from the procedure when possible; thin forks only when biology differs
CI smoke + test	`workflow_engine/domain/checks/<pack>/` + procedure tests
Usage guide	`docs/usage/*-pack.md` describing only what is instance-specific
Regulatory roadmap row	Status under Regulatory-Ready Platform

Committed stubs and examples live under docs/examples/samd/.

Reused programs and profiles¶

Default methylation path with a procedure:

sample_prep*.program.json          ← procedure.samplePrepProgram
  → study_validation_lifecycle*.program.json  ← procedure.lifecycleProgram
  → samd_research (+ researchMode from procedure) → holdout → pivotal

When a lifecycle node is biologically wrong (for example Houseman/HiTIMED on plant tissue or default cfDNA plasma), the procedure points at a thin program fork that drops that node—still not a new process pack. See plant abiotic stress and study_validation_lifecycle_no_deconv.program.json.

SaMD ladder SOP: ch.18. Agricultural or other non-clinical packs may stay on samd_research without climbing pivotal claim tiers.

Overlay knobs¶

Key	Typical use
`pipelineProcedure`	Assay recipe (`buffy_wgbs_pangenome_gene_fc`, `cfdna_wgbs_plasma`, …)
`regulatory.primary_modality`	Keep `methylation` for these packs
`regulatory.primary_analyte`	Must match procedure `analyteExpectation` — see ANALYTE_PROFILES
`actionConfig.mapper.disease_term` / `enrich_disease`	Human disease priors (OpenTargets); set `enrich_disease: false` for non-disease traits
`actionConfig.enricher.library_preset`	e.g. `neuro-core`, `plant-stress-core` (overrides analyte default such as `cancer-core`)
`actionConfig.enricher.organism` / `string_species`	Non-human Enrichr / STRING taxon
`runProgressionAnalysis` / `actionConfig.progression`	Staged indications only
`actionConfig.validation.holdout_*`	Locked holdout partition for SaMD enrichment / pivotal
`actionConfig.methyl_extract.target_panel_bed`	EM-Seq / hybrid-capture panel BED (study or site asset)

Merge precedence (highest wins first): instance → procedure → profile/mode → analyte → site (config-not-code).

Checklist for a new application pack¶

Choose a procedure (pipelineProcedure) that matches analyte and assay recipe.
Choose binary vs staged comparisons; set chromosomes / contexts for the species.
Scaffold on /work with methyl-study-init (or copy an example under docs/examples/samd/).
Author the context overlay (procedure id + preset + disease/trait knobs + progression).
Add or reuse an enrichment preset via methyl-cfg sync-library-presets when needed.
Confirm analyte + site: human GRCh38 vs plant/other linear genome pins; for EM-Seq set target_panel_bed. For epi-GBS use libraryProtocol: epi_gbs + sample_prep_epigbs.program.json (or a future epi-GBS procedure).
Point methyl-workflow-run at the procedure’s lifecycle program + overlay.
After a baseline succeeds, tune Tier-A validation knobs with hyperparameter search on that same procedure — do not put pipelineProcedure in the grid (ch.15).
Add CI smoke fixture + pack regression test (mirror Alzheimer / plant tests).
Write a short usage chapter (instance-only deltas) and add a regulatory roadmap row.
Validate: methyl-study-validate-manifest --profile samd_research (or higher ladder tier).

What an application pack does not add¶

A new omics modality or aligner (that is a process pack).
A full SamplePrep/covariate recipe (that is an assay procedure pack).
Hard-coded disease or crop names in Python — stay in manifests, overlays, and site pins.
Clinical-performance claims without the SaMD partition ladder (ch.18).