Skip to content

Regulatory-Ready Platform for Multiomics Diagnostics

If MethylPipeline were packaged and sold as a commercial product, its positioning would bridge bioinformatics R&D engineering and regulatory / quality system readiness for diagnostic development.

Moving a biomarker or multiomics discovery from laboratory R&D toward FDA authorization (commonly 510(k), De Novo, or—when required—PMA, plus laboratory frameworks such as CLIA / LDT practice) is historically slow, vulnerable to validation errors (for example train/holdout leakage), and bottlenecked by documentation audits. MethylPipeline is built to reduce those failure modes with config-first science, enforced partition discipline, and submission-oriented evidence scaffolds.

Claim boundary: The platform enforces process and configuration controls and assembles evidence packages. Architecture, SaMD profiles, and filled feasibility packages are not FDA clearance or approval. See samd-submission-scaffold.md.

1. Core value proposition

Position MethylPipeline as the "Regulatory-Ready Platform for Multiomics Diagnostics." Messaging rests on three pillars:

  • R&D agility without coding chaos. Scientists test new cohorts, modeling backends (ECDF, tabular scikit-learn, covariates), and cell-type estimation (Houseman / HiTIMED) through a configuration-first surface (config-parameter-matrix.md), not ad hoc scripts. Named assay procedure packs (pipelineProcedure) give one-shot recipes for buffy coat, cfDNA WGBS, and EM-Seq without hand-merging aligner, informME, deconvolution, and FeatureCuts knobs.
  • Production-scale cloud efficiency. Cluster-parallel execution and the Content-Addressed Action Store (CAAS) skip redundant work when only downstream steps change, reducing VM cost on large cohorts.
  • SaMD-oriented process controls. Code and profiles guard against statistical contamination (for example mixing development and locked holdout patients), stage clinical-performance claims, and help auto-assemble regulatory submission scaffolds (samd-submission-scaffold.md).

Product packaging layers

Buyers purchase a platform plus a growing catalog of config packs—not a single hard-coded assay:

Layer Owns Operator picks
Process pack Omics modality (actions, DomainPrograms, QC) regulatory.primary_modality (methylation, rnaseq, proteomics)
Assay procedure pack How the matrix is sequenced and scored (library protocol, SamplePrep/lifecycle, gene FeatureCuts, covariates) pipelineProcedure (e.g. buffy_wgbs_pangenome_gene_fc)
Application pack Indication or trait (cohorts, partitions, disease/trait overlay, enrichment preset) Study manifest + context_*.json

Merge precedence stays config-not-code: instance → procedure → profile/mode → analyte → site. Guide: Methylation application packs (procedure JSON under workflow_engine/domain/profiles/procedures/).

flowchart TD
  process[ProcessPack_methylation]
  procBuffy[Procedure_buffy_pangenome_gene_fc]
  procCfWGBS[Procedure_cfdna_wgbs_plasma]
  procEmseq[Procedure_cfdna_emseq_targeted]
  appPCa[App_oncology]
  appAD[App_Alzheimer]
  study[Study_manifest]

  process --> procBuffy
  process --> procCfWGBS
  process --> procEmseq
  procBuffy --> appPCa
  procCfWGBS --> appPCa
  procCfWGBS --> appAD
  procEmseq --> appPCa
  appPCa --> study
  appAD --> study

Analyte and disease-process roadmap

The execution platform (DomainProgram, typed actions, cfg/wf, portal, gateway workers, CAAS) is disease- and analyte-agnostic. Shipped science packs today are strongest on DNA methylation (buffy coat and cfDNA), including oncology cohorts. Near-term packs use the same control plane:

Pack Status Notes
Methylation process (WGBS / EM-Seq) In production use SamplePrep → MC stability → freeze → model; SaMD ladder.
Assay procedures — buffy_wgbs_pangenome_gene_fc In production use Default human buffy research: WGBS pangenome (methylGrapher dual C2T/G2A, alignmentMode: pangenome_wgbs), read-level informME, Houseman deconv, gene FeatureCuts. Linear alternate: buffy_wgbs_linear_gene_fc. Stock Giraffe (pangenome) is a separate non-WGBS path.
Assay procedures — cfdna_wgbs_plasma In production use Plasma WGBS + fragmentomics, gene FeatureCuts, no cell-deconv lifecycle (study_validation_lifecycle_no_deconv).
Assay procedures — cfdna_emseq_targeted Shipped (research) Inch-wide / mile-deep: libraryProtocol: emseq_targeted, sample_prep_emseq, operator panel BED (methyl_extract.target_panel_bed), elevated min_cov, no deconv.
Assay procedures — plant_wgbs_gene_fc Shipped (research) Plant WGBS trait recipe; pairs with plant_tissue analyte and plant lifecycle.
RNA-Seq process Shipped (research) primary_modality: rnaseq. Parabricks rna_fq2bam (STAR) or kallisto via actionConfig.rna_align.quant_mode. Typed actions, RNA QC, expression contract, DE gene-panel + tabular classification. Gate: representative cohorts + validation evidence.
Proteomics process Shipped (research) primary_modality: proteomics. GPU DIA-NN; CPU DDA via Sage; panel-matrix ingest; Prosit / Casanovo. Shares samples x features seam with RNA-Seq. See ch.22.
Alzheimer cfDNA application Shipped (research) Disease overlay on cfdna_wgbs_plasma: staged Control → MCI → AD, neuro-core, progression. No new actions. See ch.21.
Prostate cancer application (App_oncology) Options catalog shipped (examples pending) Multi-path operator catalog: buffy ~30× WGBS, plasma ~30× WGBS, EM-Seq targeted, and tumor-tissue progression designs. Usage ch.25. Physician/payer narrative: Prostate Cancer Application Deep-Dive. docs/examples/samd/prostate-* stubs are still a follow-on — pattern in ch.24.
Plant abiotic stress application Shipped (research) Trait overlay on plant_wgbs_gene_fc: Control vs Drought, multi-crop sites, plant-stress-core, offline plant_traits prior. See ch.23.

See RNA-Seq process pack, Proteomics process pack, and Methylation application packs.

2. Product packaging and licensing models

To capture early R&D mindshare and high-value commercial contracts, adopt an open-core model with a clear license boundary.

graph TD
    A[MethylPipeline Core] -->|Open Source AGPL-3.0| B[Community Edition]
    A -->|Commercial License + Enterprise Support| C[Enterprise SaMD Edition]
    B -->|Audience| D[Academic Research, Consortia, Devs]
    C -->|Audience| E[Biotech Startups, Pharma, Reference Labs]
    C -->|Add-ons| F[EpiPortal UI, Secure Vault Secrets, FDA Evidence Index Tool]

A. Community Edition (open core)

  • Model: Open source under AGPL-3.0 (copyleft for network/SaaS use; drives upgrade when vendors productize without contributing).
  • Included: Local workflow engine (LocalWorkflowEngine), DNA methylation process pack, shipped assay procedure packs (buffy / cfDNA WGBS / EM-Seq targeted / plant), basic CLI tools, and research-oriented profiles such as samd_research.
  • Goal: Citations, academic adoption, and developer contributions.

B. Enterprise / SaMD Edition (commercial license)

  • Model: Annual subscription (per active worker node or named deployment site), plus support. Optional per-assay / per-submission packaging for regulated buyers.
  • Included:
  • Regulatory suite: Assisted generation and packaging of validation-evidence-index.md and traceability-matrix.md for locked studies.
  • Pivotal SaMD ladder unlocks: Operator support and tooling around samd_holdout_enrichment and samd_pivotal (locked holdouts, claim gates) as described in samd-submission-scaffold.md.
  • Multi-tenant control plane (EpiPortal UI): Web UI for laboratory operators, Gantt-style run supervision, and admin workflows—not a direct SQL console for day-2 ops (portal UI → Azure SQL; workers → gateway REST only; see production-platform.md).
  • Enterprise security: Zero-trust credential distribution (cfg.credential in the DB; secrets expanded into task payloads over TLS—never as plain files under /work).
  • Hardware acceleration: Supported NVIDIA Parabricks GPU paths for alignment-heavy SamplePrep (linear fq2bam, stock pangenome Giraffe, and methylGrapher WGBS pangenome for buffy procedures).

C. Hybrid cloud (SaaS control plane + BYOC)

Genomic data is sensitive (HIPAA, GDPR). Offer Bring Your Own Cloud (BYOC): host the control plane (EpiPortal + DB metadata / scheduler); customers run stateless workers against their own AWS/Azure (or other) cluster where /work and samples reside.

3. Go-to-market (GTM) strategy

Who is the buyer?

  1. Biotech and diagnostic startups (primary). Early-detection liquid biopsy and related methylation (or upcoming multiomics) programs that need a ready QA / regulatory framework without building a platform from scratch—pick a procedure (cfdna_wgbs_plasma or cfdna_emseq_targeted) then an indication overlay.
  2. Pharma / clinical trial sponsors. Multi-cohort longitudinal studies using methylation (or additional omics packs) as endpoints; need locked, reproducible workflows over multi-year periods.
  3. Clinical reference labs / CROs. Sequencing-as-a-service providers who want to upsell FASTQs into curated classification and evidence packages—including oncology procedures and the Alzheimer cfDNA application pack as they ship.

Customer journey / “hook”

  • Research and feasibility (free / low cost). R&D uses Community Edition to process samples under a named procedure + samd_research and discover stable gene (or panel) features.
  • The “FDA chasm” (commercial trigger). When a promising marker must move to internal and pivotal validation—locked pipeline, holdout discipline, and an auditor-facing evidence package—the organization upgrades to Enterprise / SaMD for pivotal ladder support, portal operations, and regulatory packaging (samd-submission-scaffold.md).

Pathway messaging should stay honest: novel diagnostics may require De Novo or PMA rather than a predicate 510(k); laboratory deployment may also implicate CLIA. The product shortens the software and validation process path; it does not choose or guarantee the regulatory pathway.

4. Codebase-backed selling points

For technical stakeholders (CTOs, lead bioinformatics engineers), point at architecture that already exists:

  • Audit trails with low overhead. .action_results records input/output hashes, CLI versions, and environment signatures—a traceability / provenance chain suitable for quality review.
  • Cost savings via CAAS. Changing only a downstream classification step can skip alignment and extraction across large cohorts (CAAS).
  • Three-layer extensibility. New process packs add DomainPrograms and typed actions on the same scheduler (domain-program-language.md). New assay procedure packs are versioned JSON recipes (pipelineProcedure)— library protocol, SamplePrep/lifecycle pointers, FeatureCuts axis, covariate stack—without disease names in Python. New application packs reuse a process + procedure with config only (manifest, overlay, partitions, optional preset); see Methylation application packs.
  • Shipped multiomics process packs. RNA-Seq reuses the Parabricks GPU SamplePrep pattern (sample.parabricks_rna_fq2bam / sample.kallisto), adds RNA QC and an expression contract, and feeds DE gene selection into the same tabular classifier and covariate stacking. Proteomics adds GPU mass-spec ingest (DIA-NN, Prosit, Casanovo) plus CPU panel ingest, sharing a generalized samples x features seam with RNA-Seq.
  • Shipped methylation applications. Alzheimer cfDNA declares pipelineProcedure: cfdna_wgbs_plasma plus a staged Control → MCI → AD overlay (neuro-core). Plant abiotic stress declares plant_wgbs_gene_fc plus a species-agnostic trait overlay (plant_tissue, multi-crop Ensembl Plants sites, plant-stress-core, offline plant_traits prior — Open Targets remains human-only).
Document Role
Regulatory folder README Index of regulatory synthesis docs
Product and operational controls Feature inventory and control summary
SaMD submission scaffold 510(k) / De Novo-style content map
Production platform Day-2 access: portal UI + worker gateway
SaMD study lifecycle (usage ch.18) Operator SOP for the profile ladder
Methylation application packs (usage ch.24) Process vs procedure vs application
Assay procedure packs plan Implementation record for pipelineProcedure