Platform
One virtual-patient engine under six products
Sanenza is a platform with apps on top. Every product shares the same patients, the same surrogate models and the same record of how well past predictions matched reality.
Run run_7f3a · NSCLC dose comparison
- Cohort
- chr-nsclc-eu-v3
- Virtual patients
- 2 × 10,000
- Surrogate
- pkpd-onc-0.6
- Outputs
- ORR · grade 3+ AE · AUC
- Evidra record
- ev_20a1
- Run time (target)
- under 1 hour
Architecture
Six layers, one data model
| Layer | What it does | Built on | Products |
|---|---|---|---|
| Data | Ingests omics, EHR, imaging features, prior-trial and real-world data under your agreements. | RAPIDS, Parabricks, confidential computing | All |
| Virtual patients | Generates calibrated cohorts with coherent physiology, genomics and comorbidity profiles. | Conditional generative models, Clara and MONAI phenotypes | Cohorta, Physova |
| Simulation | Runs drug scenarios across the cohort and returns efficacy and safety distributions. | Modulus surrogates, BioNeMo response, CUDA, TensorRT | Insilon, Vigila |
| Design | Searches the protocol space and drafts the reasoning behind each ranked design. | NeMo agents with Guardrails, RAG over the outcome graph | Protova |
| Validation | Back-tests predictions against realised outcomes and keeps an immutable record. | cuGraph outcome graph, versioned ledger store | Evidra |
| Feedback | Sends realised outcomes back to recalibrate cohorts and surrogates. | Calibration models, model registry | All |
The flywheel
Every programme makes the next prediction better
Predict
A simulation predicts response and adverse events for a design.
Observe
The real trial or real-world outcome arrives.
Compare
Evidra links the two and scores calibration.
Recalibrate
Cohort and surrogate models update for the next run.
Deployment
Run it where your data is allowed to live
Cloud
Sanenza Cloud
Managed, single-tenant, in US and EU regions. Fastest way to start a pilot.
Your cloud
Private deployment
The platform runs in your cloud account on confidential-computing GPUs. Planned for 2027.
On-premises
Secure data centre
For sponsors who cannot move data. DGX-class servers on site. Planned after the first pilots.
Governance
Every number can be traced to its inputs
Lineage
Versioned cohorts and models
Each result names the cohort version, surrogate version, seed and data snapshot that produced it.
Review
People approve, agents propose
Protova proposes designs and drafts text. A named reviewer approves anything that leaves the platform.
Audit
Immutable records
Inputs, runs, approvals and outcome comparisons are kept in Evidra and can be exported for inspection.
Integrations
Connects to the tools your modellers already use
Connectors marked planned are being designed with pilot partners.
| Source or target | Direction | How | Status |
|---|---|---|---|
| Trial data (SDTM, ADaM) | In | CDISC datasets by SFTP or object storage | Planned |
| Real-world data and EHR | In | OMOP CDM extracts under your data agreements | Planned |
| Omics | In | VCF, FASTQ through Parabricks pipelines | Planned |
| Imaging features | In | DICOM-derived features from Clara and MONAI | Planned |
| NONMEM, Monolix, R, Python | Out | Parameter and simulation exports, Python client | Planned |
| SAS and statistical packages | Out | CDISC-aligned result tables | Planned |
| Document systems | Out | Evidence packages as PDF and structured JSON | Planned |
See the platform on your own question
A briefing shows your team the platform on a public example. A pilot then connects one programme's data, back-tests on known outcomes and runs one live decision.