λ INFERENCE FIELD
MODE / LATENT ATLAS
The Materials That Should Exist
AI, holograms and biology converge in hyperdimensional design spaces to create next-generation biomaterials for better care.
Imagine designing a biomaterial before it exists.
A biomaterial is matter in conversation with life. It carries chemistry, structure, cells, signals, fluids, tissues and time in the same system. In our lab, biology and material are part of the same design workflow from the first act of imagination.
We design engines, models and holograms using real-world material and biological data to generate hyperdimensional spaces beyond the reach of our eyes, but not beyond imagination. Our vision is to create next-generation biomaterials for better care through unthinkable and unimagined design strategies.
We don't improve the materials that exist. We design the ones that should.
Mapping the variables that control material behaviour.
A scientist can test one combination at a time. The platform maps the formulation space, ranks the variables that matter, and identifies the experiments most likely to change the answer. The scientist defines the question; the model narrows the search; the lab validates the answer.
Test everything — or test what matters.
Say you want to map how a calcium-silicate cement's recipe drives two things at once — how fast it sets and how strong it gets. You can vary several factors, each at several levels — the actual values you test. Pick what to study and how finely, watch the experiments multiply, then let an orthogonal array fold them into a fraction of the runs that still answers the question.
Turning material structure into design decisions.
The platform links what a material is made of to how it performs. We quantify composition, porosity, fibre architecture, interfaces and biological response, then use those data to predict and redesign material performance.
The model learns a material from its data: composition, microscopic structure, porosity, fibre networks and how it behaves over hours and days. From those patterns it can predict properties it has never measured, and propose the formulation that would reach a target.
Some variables drive performance. Most only add noise.
Imagine you must optimise one of the materials below with as few experiments as possible — and you already know the target you're aiming for. The instinct is to measure everything. But for any given target only a handful of factors truly move it; the rest barely matter — a scaffold's strand count changes its stiffness, yet hardly touches how fast cells multiply. The skill is spotting which to measure and which to skip. Pick an application, switch factors on one at a time, and watch the model's confidence climb toward the target — until adding more stops paying off.
Pick an application above, then turn factors on to feed the model.
Use early chemistry to predict long-term behaviour.
As a calcium-silicate cement sets it releases hydroxyl ions, and the pH around it climbs. That rising alkalinity is what makes it bioactive: too little and the material does nothing; too much and it stresses cells. Set the alkalinity you're aiming for, then act as the model. Feed it the first readings (pH at 3 h and 24 h) plus the specimen size, and watch it forecast the full 28-day curve. Then check it against real measurements.
The alkalinity you want at 28 days. The shaded band is the bioactive window — aim inside it.
Evolving formulations toward a target profile.
Genetic algorithms search competing formulations and converge on recipes that best match the target properties. We begin with many candidate cement recipes; each disc is one mix of components. Generation after generation, selection lets the unfit fade into dead-ends and the population converges toward the target you set.
The catch: trade-offs. Some demands pull against each other. Strength ↔ radiopacity: the dense radiopacifier that makes a cement show on X-rays also weakens it. Flowability ↔ set speed: the water that thins the paste also makes it set more slowly. No single recipe can maximise everything at once, so evolution has to find the best compromise. Set the target, then press Play.
Set the demands, then press Play to watch evolution design the recipe.
A closed-loop platform for biomaterials design.
Five capabilities run as one closed loop. Each stage feeds the next, and validated evidence circles back to sharpen the model — the engine learns as it works.
05 STAGES
Screen
A compact design finds which factors actually govern the material.
Infer
Models forecast behaviour and rank candidates before the bench.
Design
Given a target, the engine prescribes the recipe that delivers it.
Confirm
Predictions are tested experimentally in material assays and cell models before they are treated as evidence.
Apply
Validated findings move toward regenerative, clinical reality.
A material is not biocompatible because one assay says so.
A material can help cells multiply and still fail to let them move, or to rebuild bone. Judge it on one measure and you are fooled. The Digital BioScore integrates proliferation, migration and mineralization into a single biological profile, so materials are judged by coordinated tissue response rather than one isolated endpoint. Like the planes of a hologram, the domains combine into a single read, scored high, mid or low.
Toward biomaterials matched to the patient, the tissue and the clinical task.
Within five years, the loop closes around the clinic: prediction that reads an individual case, and discovery that runs on its own, with the lab proposing, testing and learning with less and less human hand-holding. The long-term goal is to move beyond average-case materials toward formulations selected for the biology and constraints of each case.
Models that read the individual case and forecast how a material will perform for that person — not the average patient.
A self-driving loop that proposes, tests and refines biomaterials on its own — the scientist setting direction, the machine doing the search.
Evidence from the field.
Three studies show the same principle in action: fewer experiments, earlier prediction, targeted formulation and experimental validation. Open any to read the full story.
The field is open.
Human judgement sets the biological question. AI narrows the design space. Experiments decide what is true.