21 alzheimer cfdna pack

Alzheimer detection ships as an application pack (disease) on the existing DNA-methylation process pack, not a new omics modality. The generic pattern— manifest, overlay, partitions, optional enrichment preset—is documented in Methylation application packs. This instance declares assay procedure cfdna_wgbs_plasma (linear WGBS, fragmentomics, gene FeatureCuts, no cell deconvolution), the samd_* profile ladder, and the cfDNA analyte profile; the Alzheimer-specific parts are cohorts, patient-disjoint partitions, and a small disease overlay. There is no new aligner, action, or program.

For the shared per-sample and Monte Carlo topology, see the DNA methylation end-to-end workflow. This chapter covers only what is Alzheimer-specific.

Committed example: docs/examples/samd/alzheimer-cfdna/.

Cohort design (staged Control -> MCI -> AD)¶

The study manifest expresses two ordered disease stages under one group, giving two comparisons plus progression:

flowchart TD
  H["healthy.csv (Control)"] --> CMP{"comparisons control_vs_each_disease"}
  MCI["mci.csv (AD_MCI)"] --> CMP
  AD["ad.csv (AD_AD)"] --> CMP
  CMP --> MC["MC stability per comparison"]
  MC --> PROG["pipeline.progression progression_labels: AD_MCI, AD_AD"]
  PROG --> MODEL["freeze + model + holdout (SaMD ladder)"]

Alzheimer cfDNA pack cohort flow

Manifest essentials (project_Healthy_vs_AD_Stages.json):

{
  "diseases": {
    "label": "AD",
    "groups": [{
      "label": "AD",
      "stages": [
        {"label": "MCI", "sample_paths": ["data/mci.csv"]},
        {"label": "AD",  "sample_paths": ["data/ad.csv"]}
      ]
    }]
  },
  "comparisons": "control_vs_each_disease",
  "progression_order": "explicit",
  "progression_labels": ["AD_MCI", "AD_AD"],
  "regulatory": { "primary_modality": "methylation", "primary_analyte": "cfdna" }
}

regulatory.primary_modality: methylation keeps this on the methylation process; primary_analyte: cfdna selects the cfDNA analyte profile (fragmentomics + alignment guardrails). For a binary study instead, use a single disease group and drop the progression block.

Disease overlay¶

Alzheimer biology is expressed as an instance/context overlay merged onto procedure cfdna_wgbs_plasma + the samd_* profile (context_alzheimer_cfdna.json):

Key	Value	Why
`pipelineProcedure`	`"cfdna_wgbs_plasma"`	Assay recipe: plasma WGBS, no cell deconv lifecycle, gene FeatureCuts.
`actionConfig.mapper.disease_term`	`"Alzheimer's disease"`	OpenTargets disease prior; writes `disease_associated` columns. Overrides the mapper's prostate CLI default.
`actionConfig.mapper.enrich_disease`	`true`	Enable the disease-gene annotation.
`actionConfig.enricher.library_preset`	`"neuro-core"`	Brain/CNS gene sets; overrides the cfDNA analyte default `cancer-core`.
`runProgressionAnalysis`	`true`	Ordered MCI -> AD progression scoring.

neuro-core enrichment preset¶

neuro-core is a committed enrichment library preset in the preset registry (packages/methylenricher/methyl_enricher/data/library_presets.json), synced into the cfg registry like the action catalog:

methyl-cfg sync-library-presets      # registry -> cfg (kind enrichment_library_preset)
methyl-cfg materialize               # -> /work/site/enrichment/*.json

It keeps the pathway/TF/disease-prior libraries and adds brain/CNS and aging gene sets (e.g. Allen_Brain_Atlas_up/down, Descartes_Cell_Types_and_Tissue_2021, Aging_Perturbations_from_GEO_*) while dropping oncology-only drug libraries. See enrichment library presets.

SaMD ladder¶

The pack climbs the same claim ladder as any methylation study (SaMD study lifecycle):

Tier	Profile	Required partition
Research	`samd_research`	`locked_test` recommended
Enrichment	`samd_holdout_enrichment`	non-empty `locked_test`
Pivotal	`samd_pivotal`	non-empty `pivotal_validation`

methyl-study-validate-manifest \
  --project /work/projects/alzheimer-cfdna/configs/project_Healthy_vs_AD_Stages.json \
  --profile samd_holdout_enrichment

methyl-workflow-run \
  --program workflow_engine/domain/fixtures/study_validation_lifecycle_no_deconv.program.json \
  --context-file /work/projects/alzheimer-cfdna/configs/context_alzheimer_cfdna.json

Clinical-performance claims stay blocked until regulatory.stage = pivotal_validation with a non-empty pivotal cohort.

What this pack does not add¶

No brain/CNS cell-type deconvolution basis (shipped bases are blood-only). The cfdna_wgbs_plasma procedure omits pipeline.cell_deconvolution by default; optional HiTIMED plasma overlays remain instance-level advanced knobs.
No new DomainProgram, action, or aligner. Only cohorts, partitions, the overlay, and the neuro-core preset.