What Are Digital Twins of Humans — and Why Does Medicine Need Them Now?
Digital twins of humans are physics-based, AI-powered virtual models that replicate how a real person's body moves, responds, and behaves. They are built from real-world data — biometric readings, medical imaging, motion capture, and more. And right now, a New York-based startup called Mantis Biotech is using them to tackle one of medicine's most stubborn problems: the lack of reliable data for rare diseases, edge cases, and complex human conditions. The solution they have built is unlike anything the biomedical world has seen before.
 |
| Credit: Google |
The Data Problem That Has Been Holding Medical AI Back
Anyone working in biomedical research knows the frustration intimately. Artificial intelligence thrives on data — the more, the better. But rare diseases, by definition, affect very few people. That means very little data. Regulatory and ethical constraints make it even harder to access what little does exist. Patient privacy laws, institutional silos, and inconsistent data formats all combine to create a landscape where AI models fail exactly when they are needed most.
This is not just a technical inconvenience. It has real consequences for real patients. Surgical robots need training data to be accurate. Predictive models for patient outcomes need representative samples across body types and conditions. Clinical trials need reliable baselines to measure how patients respond to treatments. When that data does not exist — or simply cannot be accessed — medical innovation slows to a crawl. Mantis Biotech was built from the ground up to solve this exact problem.
How Mantis Biotech Actually Builds a Digital Twin
The Mantis platform works by pulling together data from wildly different sources — medical imaging scans, biometric sensor readings, training logs, motion capture footage, and even academic textbooks. An LLM-based system then routes, validates, and synthesizes all of those data streams into a single coherent model. That model is then processed through a physics engine, which produces what Mantis calls a high-fidelity render — a realistic, physics-grounded simulation of a living human body.
The physics engine layer is what makes this approach genuinely different from conventional synthetic data generation. By anchoring every output to real physical laws — how muscles contract, how joints absorb force, how the body compensates for injury — the platform can generate realistic synthetic data even for scenarios where no real-world dataset exists. Mantis Biotech founder and CEO Georgia Witchel offered a vivid example: hand-pose estimation for a person missing a finger. No publicly available labeled dataset for that scenario exists anywhere. With the Mantis platform, generating it is straightforward — remove the finger from the physics model and regenerate. It takes minutes, not years.
Why the Physics Engine Is the Real Breakthrough Here
Most synthetic data platforms generate information that looks statistically plausible on paper. Mantis goes further — and that difference is everything. Because the physics engine understands anatomy and biomechanics at a fundamental level, the synthetic data it produces is not just statistically similar to real data. It is physically realistic. That distinction matters enormously for medical and scientific applications, where a model that behaves incorrectly in physical terms could lead researchers — and clinicians — to dangerous conclusions.
This also means the platform can safely explore conditions that would be impossible or deeply unethical to study directly on human subjects. Witchel describes the ideal mindset for working with digital twins as the same uninhibited energy a toddler brings to a toy — testing, stressing, experimenting without hesitation — because the subjects are entirely virtual. No patient is harmed. No consent form is required. No ethics review board needs to sign off on every simulation. The privacy of real individuals is not just protected — it is rendered irrelevant.
Digital Twins in Professional Sports — Where the Technology Is Already Working
Mantis Biotech has found its early commercial footing in professional sports, and that makes a lot of sense. Elite sports teams have both the data infrastructure and the financial incentive to invest in predictive human modeling. The startup counts an NBA team among its primary clients, using digital twins to track how individual athletes jump, move, and perform — not just on game day, but continuously across an entire season.
The platform correlates performance changes with variables like sleep quality, training volume, and upper-body movement frequency. Coaches and medical staff get an unprecedented window into each athlete's condition, long before a problem becomes visible on the court. The use cases extend well beyond basketball too. Witchel described how a football team could use the same technology to predict the likelihood of a specific player developing an Achilles tendon injury, modeled against their performance data, training load, dietary habits, and career history. The system does not just describe what has already happened — it anticipates what is likely to happen next.
Solving Rare Disease Research Through Synthetic Human Data
The most consequential application of this technology may not be in sports arenas at all. It may be in research laboratories. Rare diseases represent a persistent and painful blind spot for medical AI. Because patient populations are small, data collection is difficult, and the ethical constraints around accessing patient data are strict and necessary, these conditions are chronically under-researched relative to how devastating they can be for those who live with them.
Digital twins offer a genuine path forward. The Mantis platform can generate synthetic patient datasets that reflect the physiological realities of rare conditions, allowing researchers to run experiments, train models, and test hypotheses without ever needing access to real patients. This is not about replacing human data with inferior substitutes — it is about generating physics-grounded virtual equivalents that are rigorous enough to be scientifically useful while protecting real individuals completely. Witchel is emphatic on this point: the goal is not to exploit people's data differently. It is to remove the need for that data entirely.
FDA Trials, Pharmaceutical Research, and the Road Ahead
Mantis Biotech recently closed a 7.4 million dollar seed funding round, led by Decibel VC and supported by Y Combinator, Liquid 2, and a group of angel investors. The capital will fund hiring, marketing, and go-to-market expansion as the company prepares to move beyond its current client base. But the product roadmap reveals the company's real ambitions.
The next phase involves releasing the platform for broader use in preventative healthcare — helping individuals and clinicians understand risk before disease ever takes hold. At the same time, Mantis is actively working with pharmaceutical labs and researchers running FDA trials to model how specific patient populations respond to new treatments. If that effort delivers on its promise, it could meaningfully shorten the timeline from drug discovery to approval, and provide regulators with richer, more reliable evidence than traditional trial designs alone can generate.
A New Layer of Infrastructure Beneath All of Medical AI
What Mantis Biotech is really building is not just a product — it is infrastructure. A foundational layer beneath all the AI models that the rest of the healthcare world is desperately trying to deploy. Those models are only as good as the data they are trained on. By generating physically realistic synthetic data for any human condition, any body type, any rare or unusual scenario, Mantis addresses the single most fundamental constraint that has limited medical AI from the very beginning.
The physics engines already exist. The large language models already exist. The biometric data collection infrastructure already exists. What Mantis has done is combine these elements in a novel and purposeful way, laser-focused on filling the gaps that conventional data collection cannot reach. The implications span every corner of biomedicine — from surgical robot training to drug development timelines, from elite sports performance optimization to treatments for diseases that currently have almost no research base at all.
Digital twins of humans have arrived. And if Mantis Biotech's early momentum is any signal of what comes next, they may be about to change what medicine knows about the human body — and what it can finally do with that knowledge.
