Virtual Cell · evidence check

Ask the disease question first; keep benchmark evidence separate.

Define disease, cell state, perturbation, target, time and endpoint. NMD-VCell returns the query verdict on the left and any reusable benchmark reference on the right, so an aggregate HepG2 result is never mistaken for a target-specific DMD prediction.

Run evidence check Inspect supporting data

Current value: the platform identifies exactly why a DMD virtual-cell response is not yet estimable and turns that gap into a decisive experiment.

NMDVirtual
Cellevidence-aware

Cell state Perturbation Time Phenotype Uncertainty

Module meaning contract

Virtual Cell Studio · what this module can change

See all eight modules →

Question answeredCan the current release build this cell context, execute the perturbation and return a governed state transition?

Scope & evidence unitThe interactive six-input predictability checker is DMD-specific. FSHD, DM1 and SMA currently expose bounded evidence routes, not an inference engine.One question object with disease, cell state, perturbation, time, endpoint and requested output fixed together.

ProducesRegistered same-assay benchmark result or explicit abstention, plus evidence layers, missing target-context outcome, claim boundary and the next discriminating experiment.

Cannot establishNo calibrated DMD response, candidate pathway outcome, patient trajectory, efficacy estimate or clinical recommendation.

Next handoffInspect the relevant context in Browse, then compare DMD candidate evidence or freeze a Study Card. Continue →

Panel contract

The selected biological query and the reusable benchmark are different objects.

Cell-state composer

Define the start state, intervention and desired readout.

Query verdict · abstention with study handoff

01Cell or state

02Disease context

03Perturbation

04Target

05Time

06Primary endpoint

The left panel answers the selected query. The right panel is a reference benchmark and must not be read as the query result.

QUERY RESULT ABSTAIN — DMD response is not currently estimable No matched candidate perturbation outcome exists for ZNF133 in a DMD-relevant muscle system.

Research objectQ:DMD:DIFFERENTIATING:CRISPRI:ZNF133:STATE:FUSION

Verdict meaningDesign-ready evidence gap

UnlocksStudy Card → DMD myogenic perturbation outcome

Machine stateOpen contract →

AssayDMD/control myogenic CRISPRiTarget engagement + fusion endpoint

ControlsMatched donors and non-targeting controlsViability and reagent concordance included

Decision ruleUpdate model eligibility after replicated outcomeEndpoint and effect margin frozen first

RegistrationFreeze the Study Card before outcome visibilityQuestion, assay and analysis boundary stay auditable

01DMD contextObserved disease evidence

→

02Muscle contextContextual myogenic evidence

→

03Candidate perturbationNo matched DMD outcome

→

04Functional endpointFusion outcome missing

MolecularMean vector2,000 features · aggregate output

Cell stateNeeds DMD truthConditioned trajectory becomes estimable after calibration

FunctionNeeds functional truthMeasure a prespecified muscle endpoint

ToxicityNeeds safety truthMeasure matched viability and toxicity

Cell–cellNeeds cross-cell truthMeasure a causal cross-cell endpoint

Next decisive experimentReplicate-aware DMD myogenic perturbation assay with molecular + fusion endpoint

Open source evidence Design the missing test Open Study Card

Computable biological assets

Four layers connect observed cells to a governed virtual state.

Inspect the complete data universe →

Layer 1 · Disease state

Measured NMD contexts

DMD patient muscle, organoid, correction, spatial and cross-species objects coexist with FSHD, DM1 and SMA modules.

Context available

Layer 2 · Perturbation outcome

Observed response substrates

2,160 HepG2 targets, 21 governed candidate responses and healthy-myoblast screening provide context-bounded perturbation outcome.

Assay bounded

Layer 3 · Model benchmarks

Registered execution records

Ridge, GEARS, scGPT, TxPert and MORPH have registered local runs. STATE remains an external architecture reference; none emits a calibrated DMD cell population.

Executed · bounded

Layer 4 · Prospective truth

DMD intervention outcomes

A matched candidate perturbation in DMD-relevant muscle is the missing calibration layer—not missing data in general.

Abstention required

State transition contract

Measured reference and desired prediction are kept visually separate.

The current platform can locate a question along myogenic time. It cannot infer the counterfactual arrow from perturbation to DMD function until matched outcomes exist.

Open trajectory evidence Open the 24-slot pilot

Measured referenceMyoblast0 h

→

Measured referenceDifferentiate24–48 h

→

Context screenFusion9/21 candidates

⇢

Desired outputDMD functionoutcomes not measured

Machine-readable product contract

The interface and its scientific ceiling are downloadable.

Use the schema to prepare a future virtual-cell run object. A valid object records modality, cell state, disease context, time, endpoint, evidence provenance and whether inference was executed.

{
  "cell_state": "proliferating_myoblast",
  "disease_context": "DMD",
  "perturbation": {"modality": "CRISPRi", "target": "ZNF133"},
  "time_hours": 48,
  "primary_endpoint": "cell_state_transition",
  "inference_status": "LOCKED_NO_DMD_OUTCOMES"
}

Open engine contract Open JSON Schema Download template