In development
Myeloblast virtual cell
Acute myeloid leukemia
Leukemic stem and blast states, their normal HSPC counterparts and the marrow niche.
CloneNormal counterpartNiche
Virtual CellsIn development
Virtual cells, in-silico experiments and drug design.
A virtual cell models the malignant clone, its normal counterpart and the niche around it, from the biobank's harmonised data. Padma runs experiments on it before the lab, and every prediction is tested back through it.
Disease-specific virtual cells
Hematologic malignancies first, where our single-cell and clonal-evolution data are deepest. Each model carries three layers: the clone, its normal counterpart and the niche.
In development
Dysplastic neutrophil virtual cell
Myelodysplastic neoplasms
Dysplastic, ineffective hematopoiesis and the clonal steps toward AML.
CloneNormal counterpartNiche
In development
Megakaryocyte virtual cell
Myeloproliferative neoplasms
JAK2-, CALR- and MPL-driven clones, megakaryocyte signalling and marrow fibrosis.
CloneNormal counterpartNiche
Next
Plasma cell virtual cell
Multiple myeloma
Malignant plasma-cell clones and the surface antigens that CAR-T and bispecific therapies target.
CloneNormal counterpartNiche
Next
B lymphocyte virtual cell
Chronic lymphocytic leukemia
B-cell receptor signalling and clonal escape under targeted therapy.
CloneNormal counterpartNiche
Reference
Stem and progenitor cell virtual cell
Healthy reference
Normal hematopoietic stem and progenitor cells: the counterpart every therapy must spare.
CloneNormal counterpartNiche
What a virtual cell holds
- The malignant clone
- Its states, drivers and dependencies, from single-cell, genomic and clonal-evolution data.
- Its normal counterpart
- The healthy cells a therapy must spare, modelled alongside the disease.
- The niche
- The microenvironment that supports the clone and shapes response.
- Calibrated uncertainty
- Every prediction carries a confidence and the data it rests on. The model abstains where evidence is absent.
Drug design on virtual cells
- Target identification
- Targets that survive independent replication and show a therapeutic window against normal tissue.
- Candidate design
- Small-molecule and biologic candidates against ranked targets; antigen selection for cell and bispecific therapies.
- Efficacy against toxicity
- Predicted effect on the malignant clone against its normal counterpart and the niche.
- Resistance
- Clonal escape modelled from clonal-evolution data: which subclones survive, and what second line follows.
- Diagnostics
- In-silico assessment of biomarker and MRD assay designs before samples are spent.
In-silico experiments: run one on a virtual cell.
Press the button. In this illustration a simulated knockout collapses the malignant clone, leaves a resistant subclone, and spares the normal counterpart: the three answers a therapeutic programme needs before wet-lab spend.
In-silico experiment
Virtual disease model, untreatedKnockout of target X, simulated
- Malignant clone
- Resistant subclone
- Normal counterpart
From data to a prediction the lab can test.
- 1
Biobank data
Genomic, transcriptomic, single-cell, proteomic and flow data for one disease, harmonised in the biobank.
- 2
Virtual cell
The malignant clone, its normal counterpart and its niche, with calibrated uncertainty on every prediction.
- 3
In-silico experiments
Padma runs knockouts and candidate therapeutics against the model before anything reaches a bench.
- 4
Stress test
Efficacy against toxicity, resistance under selective pressure, off-target effects, sensitivity to assumptions.
- 5
Back to the lab
Prioritised predictions return through the Virtual Lab for wet-lab testing by partners.
Independently evaluated before any prediction is offered
The independence firewall applies: the Trust Layer never evaluates our own models, so evaluation is commissioned from outside.
Status
In development.
Collaborations with translational and discovery teams are scoped on request. No efficacy claims are made or sold.
Also in Virtual Cells
Custom biomedical AI systems
Retrieval systems over a client’s own literature and data, and scoped agents for defined research workflows, each with a pre-registered fitness-for-purpose evaluation. Never sold as independent evaluation.
Physical AI for the laboratory
Agents that operate and audit instruments through published driver interfaces: fail-closed, human approval for high-risk steps, tamper-evident audit records. The first open-source release is live; pilots run under qualification protocols. Not sold as a validated GxP system.
Working on something adjacent?
Collaborations on virtual cells, in-silico design and laboratory automation are scoped in writing, with the independence firewall applied.