In-silico clinical trials

Prove it on virtual patients first

Sanenza simulates a drug across a realistic synthetic population, so dose, cohort and endpoint choices are tested before the first real patient is enrolled.

Sanenza is pre-launch and is looking for design partners for 2027 pilots.

Example run · compound SZ-0412 · 400 virtual patientsExample
Responders24862% of 400
Adverse events297% of 400
Rare subgroup, adverse16 of 28Low-tolerance genotype
Example population of 400 virtual patients coloured by predicted outcome at the selected dose. Illustrative values.
40 mg
  • Responder
  • Adverse event
  • No response
  • Rare subgroup

Move the dose to see the trade-off a dose-finding study has to settle.

Illustrative values from a toy model, not a real compound. A real Sanenza run uses mechanistic PK-PD surrogates, 10,000 or more patients per arm and calibrated uncertainty.

Design targets

What we are building toward

These are targets the pilot programme will measure. They are goals, not results.

Patients per arm10,000+

Design target for one simulated arm.

Protocols scored100s

Candidate designs Protova aims to compare per question.

TurnaroundHours

Target for a dose sweep that takes weeks on CPU solvers.

Pilot length10weeks

From data agreement to a back-tested report.

The problem

Trials fail on choices made too early

Most of a programme's cost is committed when the protocol is fixed. Dose, cohort and endpoints are chosen from thin, biased history, and a wrong choice shows up years later.

Dose

Picked from a handful of scenarios

Classical PK-PD and QSP models are accurate but slow, so teams explore a few doses instead of the full range.

Cohort

Built on data that misses people

Paediatric, rare-disease and under-represented groups are barely present in historical trials, so their risk stays unknown.

Safety

Rare events surface too late

An adverse event that affects one patient in a thousand rarely appears in a trial of three hundred.

How it works

A simulated trial in four steps

  1. Week 1

    Define the question

    Pick the compound, indication and decision: dose, cohort, endpoint or sample size.

  2. Week 2–3

    Build the population

    Cohorta generates a calibrated cohort from physiology priors and omic and EHR variability.

  3. Week 4–7

    Simulate and search

    Insilon runs every candidate arm; Protova scores designs on power, efficacy, safety and equity.

  4. Week 8–10

    Validate and report

    Evidra back-tests against known outcomes and writes an auditable evidence package.

Technology

Mechanistic models at GPU speed

Each virtual patient is a coupled physiology and molecular system. Running tens of thousands of them per arm needs accelerated compute.

NVIDIA BioNeMoBiomolecular target, ligand and pathway response
NVIDIA ModulusPhysics-ML surrogates for physiology and PK-PD
NVIDIA ParabricksGPU genomics that grounds patients in real omic variability
RAPIDS cuDF, cuML, cuGraphEHR, omics and outcome-graph processing
Clara and MONAIImaging-derived phenotypes and anatomy
NeMo and GuardrailsTrial-design agents and evidence drafting
Triton, TensorRT and NIMServing simulation and inference at population scale
OmniverseOrgan-scale twins in Physova
H100 confidential computingTraining and simulation on protected health data

For developers

One API from cohort to evidence

Create cohorts, start simulation runs and pull results into R, Python or SAS. Every run gets an Evidra record, so results stay traceable.

  • REST API and Python client, versioned
  • Each result carries its cohort version, model version and seed
  • CDISC-aligned exports for downstream analysis
run.sh
# Simulate two doses across a Cohorta population
curl -X POST "https://api.sanenza.com/v1/runs" \
  -H "Authorization: Bearer $SANENZA_TOKEN" \
  -d '{
    "cohort": "chr-nsclc-eu-v3",
    "compound": "SZ-0412",
    "arms": [{ "dose_mg": 20 }, { "dose_mg": 40 }],
    "patients_per_arm": 10000,
    "outputs": ["orr", "grade3_ae", "exposure_auc"]
  }'

# Response (abridged, example values)
{ "run": "run_7f3a", "status": "queued", "evidra_record": "ev_20a1" }

Questions

Common questions

Is Sanenza available today?

Not yet. Sanenza is pre-launch. We are working with a small number of design partners on 10-week pilots that start in 2027.

Does in-silico evidence replace clinical trials?

No. Simulation helps choose better doses, cohorts and endpoints, and can support model-informed drug development. Physical trials and regulatory review remain the decision point.

Do regulators accept this evidence?

Acceptance of in-silico evidence is evolving and depends on the question and therapeutic area. Sanenza produces documented, back-tested evidence for your MIDD work; it does not guarantee acceptance.

Where does the patient data come from?

From your own trial and real-world data under your agreements, plus licensed and public sources. Protected health data is processed in confidential-computing environments.

Test your next protocol on virtual patients

Start with a briefing: a Sanenza scientist walks your trial team through a back-test on a public example, then we look at one open question from your programme.