23 plant abiotic stress pack

Plant abiotic stress detection ships as an application pack (trait) on the existing DNA-methylation process pack, demonstrating that the platform is species-agnostic. The generic pattern is documented in Methylation application packs. This instance declares assay procedure plant_wgbs_gene_fc and extends the pattern with plant platform unblockers because human blood/cfDNA defaults are biologically wrong for plant WGBS: a plant_tissue analyte, a non-human site reference, a lifecycle program without blood cell deconvolution, and a plant enrichment preset. It reuses SamplePrep, the centroid → detector → mapper → Monte Carlo stability science, and samd_research. There is no new aligner or workflow action.

For the shared per-sample and Monte Carlo topology, see the DNA methylation end-to-end workflow. Plant epigenomics background (CG/CHG/CHH contexts, stress biology, MSH1) is in Plant Research. This chapter covers only what is plant-specific.

Committed example: docs/examples/samd/plant-abiotic-stress/.

Why plants are different

Plant genomes methylate cytosines in three sequence contexts, each with distinct biology, so the pack runs all three and treats non-CG methylation as signal rather than a bisulfite-conversion artifact:

Context	Maintained by	Meaning
CG	MET1	Often stable, heritable epialleles
CHG	CMT3	Chromatin / TE silencing
CHH	RdDM (24-nt siRNA)	Dynamic environmental (e.g. drought) response

This is why the human WGBS defaults (which treat CHG/CHH methylation above ~2% as a conversion failure) must be lifted for plants.

Cohort design (binary Control vs Drought)

The study manifest expresses one control group and one stress group, giving a single comparison. Grafting / trait-introgression is a separate later pack, not this one.

flowchart TD
  C["control.csv (well-watered)"] --> CMP{"comparisons control_vs_each_disease"}
  D["drought.csv (stressed)"] --> CMP
  CMP --> MC["MC stability per comparison CG / CHG / CHH"]
  MC --> MODEL["freeze + model + holdout (samd_research)"]

Plant abiotic stress pack cohort flow

Manifest essentials (project_Control_vs_Drought.json):

{
  "diseases": {
    "label": "stress",
    "groups": [{ "label": "drought", "sample_paths": ["data/drought.csv"] }]
  },
  "comparisons": "control_vs_each_disease",
  "chromosomes": ["1", "2", "3", "4", "5"],
  "contexts": ["CG", "CHG", "CHH"],
  "regulatory": { "primary_modality": "methylation", "primary_analyte": "plant_tissue" }
}

primary_modality: methylation keeps this on the methylation process; primary_analyte: plant_tissue selects the plant analyte profile (plant-safe QC, no cfDNA fragmentomics). Arabidopsis chromosomes are 1..5; swap the list for other species.

Site reference (plant genome gate)

Plants align linear only (no HPRC pangenome). Pin a non-human genome + GTF in a site manifest and provision the assets once per cluster:

scripts/download_arabidopsis_tair10.sh
export METHYL_SITE_CONFIG=workflow_engine/domain/profiles/site_tair10.example.json

The committed example site_tair10.example.json pins Arabidopsis_thaliana.TAIR10.dna.toplevel.fa and Arabidopsis_thaliana.TAIR10.58.gtf under /work/genomes and omits the pangenome, RNA and proteomics reference blocks. The genome site reference is the real deployment gate for this pack, not new Python science.

Plant analyte overlay

Plant biology is expressed through the plant_tissue analyte defaults plus a small instance/context overlay (context_plant_abiotic_stress.json):

Key	Value	Why
regulatory.primary_analyte	plant_tissue	Plant-safe QC: lifts CHG/CHH extraction caps, opens the bisulfite non-CpG cap, disables cfDNA fragmentomics (see ANALYTE_PROFILES)
actionConfig.mapper.enrich_disease	true	Offline plant gene↔trait prior (enrich_source: plant_traits); not Open Targets
actionConfig.mapper.enrich_source	plant_traits	Human Open Targets / DisGeNET stay off; join an operator TSV
actionConfig.mapper.plant_traits_path	demo drought TSV	Seed table under the example pack; replace with SoyBase / MaizeGDB / Gramene extracts for crops
actionConfig.enricher.library_preset	"plant-stress-core"	Organism-general GO terms; drops human oncology/CNS libraries
actionConfig.enricher.organism	"Arabidopsis_thaliana"	Enrichr organism
actionConfig.enricher.string_species	3702	STRING PPI taxon (Arabidopsis; 3847 soybean)
runProgressionAnalysis	false	Binary study, no ordered stages

Human-symbol Enrichr libraries do not map to Arabidopsis AGI locus IDs, so for true Arabidopsis/crop term enrichment supply a custom plant GMT via actionConfig.enricher.libraries; plant-stress-core is the discovery default. The preset is a committed enrichment library preset synced like the action catalog:

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

Lifecycle program without cell deconvolution

The pack runs plant_stress_study_lifecycle.program.json, identical to the human study_validation_lifecycle except the pipeline.cell_deconvolution node is removed (shipped deconvolution bases are human-blood only). Progression is config-gated off for the binary study.

methyl-study-validate-manifest \
  --project /work/projects/plant-abiotic-stress/configs/project_Control_vs_Drought.json \
  --profile samd_research

methyl-workflow-run \
  --program workflow_engine/domain/fixtures/plant_stress_study_lifecycle.program.json \
  --context-file /work/projects/plant-abiotic-stress/configs/context_plant_abiotic_stress.json

Other crops (soybean, maize, wheat)

Committed site recipes and study stubs ship alongside Arabidopsis (same application pack, same lifecycle program):

Crop	Site example	Download script	Manifest / overlay	STRING
Soybean Wm82	site_glycine_max_wm82.example.json	scripts/download_glycine_max_wm82.sh	project_Control_vs_Drought_soybean.json + context_soybean_drought.json	3847
Maize B73	site_zea_mays_b73.example.json	scripts/download_zea_mays_b73.sh	project_Control_vs_Drought_maize.json + context_maize_drought.json	4577
Wheat IWGSC	site_triticum_aestivum_iwgsc.example.json	scripts/download_triticum_aestivum_iwgsc.sh	project_Control_vs_Drought_wheat.json + context_wheat_drought.json	4565

scripts/download_glycine_max_wm82.sh
export METHYL_SITE_CONFIG=workflow_engine/domain/profiles/site_glycine_max_wm82.example.json
methyl-workflow-run \
  --program workflow_engine/domain/fixtures/plant_stress_study_lifecycle.program.json \
  --context-file docs/examples/samd/plant-abiotic-stress/context_soybean_drought.json

Wheat assembly is large; operators may set a chromosome subset (e.g. ["1A"]) for research runs. Pattern checklist: ch.24.

Plant cell-type deconvolution (optional)

Default lifecycle omits blood deconvolution. When a plant atlas is available:

1. Provision Houseman or HiTIMED JSON under /work (cfg roles houseman_seed_basis / hitimed_hierarchy_basis; compose HiTIMED with scripts/build_plant_hitimed_basis.py).
2. Set site actionConfig.cell_deconvolution (seed_basis_path or hierarchy_basis_path, multi-context list). plant_tissue never falls back to packaged blood atlases.
3. Run plant_stress_study_lifecycle_with_deconv.program.json.

epi-GBS SamplePrep (optional)

WGBS stays on Parabricks sample_prep.program.json. For epi-GBS set libraryProtocol: epi_gbs (see epi_gbs.profile.json) and run sample_prep_epigbs.program.json: optional sample.demultiplex → sample.docker_align → methyl QC/extract. Pin the Docker image + argv in actionConfig.docker_align. Demux offline and set demultiplex.skip: true when barcodes were already split. Details: plant-deconv-epigbs-seams plan.

What this pack does not add

• No Open Targets / DisGeNET priors (human therapeutic databases). Arabidopsis overlay uses enrich_source: plant_traits with data/arabidopsis_drought_gene_traits.tsv; crop overlays leave disease enrichment off until operators supply a crop gene↔trait table (SoyBase / MaizeGDB / WheatIS / Gramene / literature).
• No grafting / trait-introgression design — that is a planned separate pack.
• No SaMD pivotal / clinical-performance claim ladder — an agricultural research pack stays on samd_research.
• No biologically validated plant atlas in the wheel — operators supply atlas JSON.