NMD-VCell Disease Virtual Cell Platform Specify a cell-state question, inspect computability, and design the missing test Module: Documentation · evidence-to-experiment workflow NMD = neuromuscular disorders
Search Data & API Release v1.0 candidate

Current limit: the site can compare existing evidence, but no candidate has yet been independently repeated in a DMD muscle model.

Research evidence only 21 prioritized candidate perturbation records 2,160 benchmark perturbation targets 0 independent DMD replications Boundary & release
21 Workbench candidates are surfaced for consideration, not ranked as validated targets 2,160 target-level perturbation responses form the frozen HepG2 benchmark substrate No validated DMD perturbation prediction v1.0.0-database-resource Frozen 25 Jul 2026 Schema 1.1 Model ridge-safe-v2.3 Benchmark repeated-fold-v2.2 Build EA-20260730-42 DOI pending Open evidence boundary →

Competitive benchmark to implementation blueprint

Learn, borrow and apply without overclaiming.

This V94 page turns the 31 July 2026 competitive review into a machine-readable implementation contract for NMD-VCell. It records what the platform should learn from Arc, CZI, CELLxGENE, Open Targets, DepMap, VCell and CMap, and exactly what must be true before any prediction claim changes.

Open blueprint JSON Download TSV Prediction gate
Disease-specific pathCompete by DMD specificity, prospective freezing, failed-result return and object governance; competitor lessons are implementation gates, not inherited evidence.

Current maturity translation

The benchmark report becomes five execution gates.

Competitor lessons encoded6
Implementation axes5
DMD prediction unlock0
Measured outcomes0

Learn · borrow · apply

Competitor practices are translated into NMD-specific contracts.

Reference patternLearnBorrowApply to NMD-VCellAcceptance gate
Arc STATE and Virtual Cell ChallengeA model claim must be bound to task, split, hidden labels, frozen submission and permanent simple baselines.Prediction-freeze, hidden-test and baseline-retention rules.Create a DMD Prospective Benchmark where Study Card, Prediction Card and Outcome object are separate immutable records.No DMD prediction label is shown until a timestamped prediction exists before outcome access and simple baselines run on the same split.
CZI Virtual Cells PlatformData, model, benchmark and inference should use one ModelRun contract across web, CLI and local execution.Model adapter, run log, dataset digest, split manifest and reusable benchmark package.Standardize zero, train-mean, nearest-neighbour, ridge and one future population model behind a shared ModelRun object.Every run records model id, commit, container digest, data SHA, split id, seed, status, metrics and failure reason.
CELLxGENE and Vitessce-style browsersA Cell Browser must let users operate real cell objects, not only read concept cards.Dataset, donor, disease, state, batch, expression, DE, cell-proportion and spatial-neighbour filters.Upgrade Cell Browser v1 around real DMD scRNA/snRNA/spatial data with donor-level pseudobulk as the default statistical unit.At least two DMD datasets can be filtered by donor and disease, with expression, cell count, pseudobulk DE and exportable selections.
Open Targets and DepMapLong-lived platforms are entity graphs plus stable programmatic file indices, not only pages.Entity relation graph, API version metadata, file index, historical releases and provenance by source.Expose Gene, Dataset, Comparison, Study, ModelRun, Prediction and Outcome as linked objects under one release context.Each page answers: what object is studied, what evidence supports it, and which next object would change the conclusion.
VCell mechanism simulation platformMechanistic models separate model definition, parameters, solver, simulation run and result object.Solver/run/result separation and explicit parameter documentation.Keep Computability Gate, ModelRun and future mechanistic simulation objects distinct so the name virtual cell does not imply unavailable numerical simulation.UI never presents evidence checks as simulation output; future solver objects must cite equations, parameters, run logs and result files.
CLUE/CMap service continuity lessonA valuable scientific portal can lose utility if data are only reachable through an active website.Portable archive, checksum ledger, mirror plan, historical manifest and exit strategy.Keep RO-Crate, release manifest, file registry, checksums and deprecated archive policy as first-class release objects.A release can be reconstructed from manifest, checksums, schemas and archived files without relying on the interactive site.

Applied workstreams

Where this changes the platform.

IMPLEMENTED_INTERNAL_GATE

Release engineering

Health, route manifest and overlay identity are generated and checked.

Apply: Keep manifest, downloads, overlay, health and route QA as a single generated release context.

api/health api/v1.1/release_manifest.json api/v1.1/overlay_archive_identity.json api/v1.1/route_manifest.json

PARTIAL_IMPLEMENTED

Deep-link research objects

Planner can restore gene state and object trace server-side.

Apply: Extend parent inheritance from Comparison to Study Card, Prediction and Outcome object URLs.

/resource/planner?gene=ZNF133 api/v1.1/study-cards/ZNF133.json

PLANNED_REQUIRES_DATA_IMPORT

Real Cell Browser

Cell Browser is a disease ecology and computability map.

Apply: Connect real DMD single-cell, single-nucleus and spatial datasets with donor-level filters and pseudobulk DE.

api/v1.1/dataset_registry.json

PLANNED_REQUIRES_RUNNER

Executable ModelRun system

Model Cards and benchmark contracts exist; calibrated DMD model count is zero.

Apply: Run permanent baselines and future population model under one adapter and immutable ModelRun schema.

api/v1.1/model_registry.json api/v1.1/benchmark_suite.json

PLANNED_REQUIRES_EXPERIMENT_AND_WRITE_PATH

Prospective validation loop

Study Cards are drafts; predictions and measured outcomes are zero.

Apply: Create append-only Study -> Prediction -> Experiment -> Outcome chain with null, toxic, failed-QC and abstained outcomes preserved.

api/v1.1/study_card_manifest.json api/v1.1/prediction_registry.json

Research object 2.0 scaffold

The next release should promote ModelRun, Dataset and Outcome from text to contracts.

Research object envelope

  • object_id
  • object_type
  • schema_version
  • lifecycle_status
  • release_context
  • identity.parents
  • biological_context
  • perturbation
  • endpoint
  • evidence_state
  • provenance
  • files
  • validation

Dataset v2 required fields

  • accession
  • permanent_identifier
  • license
  • species
  • disease
  • donor_count
  • cell_count
  • modality
  • perturbation_type
  • endpoint
  • raw_sha256
  • processed_sha256
  • benchmark_roles
  • permitted_model_uses
  • prohibited_uses

ModelRun required fields

  • model_id
  • model_version
  • source_commit
  • checkpoint_sha256
  • container_digest
  • data_release
  • feature_schema
  • split_id
  • random_seed
  • status
  • failure_reason
  • predicted_object_ids
  • metric_object_ids
  • stdout_log_sha256
  • stderr_log_sha256

Twelve-month route

Progression is gated by evidence, not branding.

HorizonGateNMD status
0-30 daysRelease context, health, route contract, deep-link SSR and missing-data/reproducibility inventory.MOSTLY_IMPLEMENTED_IN_V93_INTERNAL_GATE
1-3 monthsDMD Cell Browser v1, Dataset Registry v2, ModelAdapter baselines and Study Card state machine.NEXT_SOFTWARE_PHASE
3-6 monthsMinimum Perturbome freeze, hidden prospective roles, Prediction Card and Outcome alpha.REQUIRES_GOVERNED_STUDY_DESIGN
6-12 monthsAt least one independent DMD Study -> Prediction -> Outcome chain with failures preserved.REQUIRES_EXPERIMENTAL_OUTCOME
Boundary. This page does not assert data-scale parity with Arc, model-platform parity with CZI, CELLxGENE-level browsing, validated DMD perturbation prediction, clinical decision support or therapeutic target ranking. v1.0 is a database and evidence-governance release; it is not a validated disease-prediction or clinical decision-support release.