Architecture specification. Figures describe the system as designed.
The reframe
A probability is not a decision.
So we do not ask will this trial succeed. We ask which of 240 designs gives this compound its best chance — and what makes the difference.
The inputs are the same ones a team already holds at the Phase I/II boundary. Nothing here requires new wet-lab work.
Illustrative. The compound is generic and the values describe the method, not a study.
Each arrow is an equation. This is what makes a result inspectable rather than asserted.
The cohort is generated, not collected. Each virtual patient carries a full parameter set drawn from published population distributions, so the spread of the simulated population matches the spread of a real one.
Illustrative. Values describe the method, not a study.
A mean response curve hides the thing that decides a trial. The sweep keeps every simulated patient separate, so the shape of the distribution stays visible: who progresses early, who responds and relapses, and how wide the middle really is.
Illustrative. A subsample is drawn; a full sweep evaluates 2.4M trajectories.
Every protocol evaluated, ordered by predicted responder rate and progression-free survival, with a recommended dose and biomarker cutoff for the design that wins.
Which patients benefit, and the mechanism that explains why they do.
Every result traced back through the biology that produced it, open to inspection by your own modelers and your regulators.
Not a score. A design.
By the time a company decides to begin a Phase II trial, it has already invested years and tens of millions of dollars. The decision still rests on limited human efficacy data. Phase I establishes safety and pharmacokinetics, while efficacy is inferred largely from animal models that translate poorly to patients. Roughly three out of four Phase II oncology programs fail, most often for lack of efficacy. Once the first patient is enrolled, the protocol is fixed and the capital is committed.
Every engineering discipline relies on simulation before committing capital. Aircraft are designed in simulators before they fly. Cars are crash-tested virtually before prototypes are built. Microchips are simulated before fabrication.
Drug development remains one of the few trillion-dollar industries still making its most expensive decisions without an equivalent capability.
Keytruda, Opdivo, and Tecentriq have generated extensive published Phase I-III efficacy, survival, and biomarker data. The model can be calibrated and checked against real clinical outcomes without generating a single new wet-lab result.
PD-1/PD-L1 is among the most extensively characterized mechanisms in oncology. We are not inventing the biology. We are making it computable at the scale of a trial design.
Immuno-oncology is one of the largest areas of pharmaceutical R&D investment, and a failed Phase II writes off tens of millions.
The simulation architecture is indication-agnostic. Extending it means replacing disease-specific biology while reusing the modeling and acceleration layers.
One validated disease model, then the next, until no pharmaceutical company designs a clinical trial without first simulating it.
Deliberately narrow, and deliberately falsifiable. We are selecting a small number of design partners for the NSCLC model.
A Phase II oncology program, either one you are designing now or one you have already completed. For a completed trial, send only what was known before it started and withhold the outcome.
A modeled design space for your compound: protocol variants ranked by predicted responder rate, a recommended dose and biomarker cutoff, the virtual cohort specification, and the causal chain behind every result. Delivered as a written report your modelers can audit line by line.
Partners shape the model's priorities, and keep everything produced for their compound.
Become a partner[Founder name]
Co-founder
[The one credential a clinical development lead would care about: mechanistic modeling, oncology clinical development, or ML for scientific computing.]
[Founder name]
Co-founder
[Same. One line, no career summary.]