Mantis Biotech, a New York-based startup, is pioneering the creation of “digital twins” – highly detailed, physics-based models of the human body – to overcome a critical bottleneck in modern biomedical research: data scarcity. While large language models (LLMs) promise breakthroughs in genomics, diagnostics, and drug discovery, their effectiveness hinges on access to comprehensive datasets. These datasets are often limited, especially when dealing with rare conditions, unusual cases, or ethically sensitive patient information.
The Challenge of Limited Data
The core issue is simple: LLMs require massive amounts of data to function effectively. But healthcare data is often fragmented, unstructured, or subject to strict privacy regulations. This leaves researchers struggling to train models on realistic scenarios, particularly in areas where data is inherently rare. For example, understanding the biomechanics of an athlete with a missing finger is difficult because labeled datasets simply don’t exist.
Mantis Biotech’s solution addresses this directly. Their platform integrates data from diverse sources – textbooks, motion capture, biometric sensors, medical imaging – and uses LLMs to synthesize complete, predictive models. The key innovation is a physics engine that grounds these synthetic datasets in realistic anatomical and physiological constraints.
How Digital Twins Work
The process involves three core steps:
- Data Integration : Gathering information from multiple sources, including structured medical records and unstructured text.
- LLM Synthesis : Using LLMs to validate, refine, and combine this data into a coherent framework.
- Physics-Based Modeling : Running the integrated data through a physics engine to create high-fidelity simulations of human anatomy and behavior.
This allows Mantis Biotech to generate synthetic datasets for scenarios where real-world data is unavailable. For example, the platform can simulate hand-pose estimation for someone with a missing finger by simply removing the digit from the model and regenerating the simulation. This bypasses the need for rare or nonexistent datasets.
Current Applications and Future Expansion
The company’s initial success is in professional sports, where high-performance athletes require detailed biomechanical analysis. One NBA team is using Mantis Biotech’s digital twins to track player performance, predict injury risks, and optimize training regimens. The platform can analyze years of jump data alongside sleep patterns, arm movements, and other biometric markers to provide actionable insights.
Mantis Biotech plans to expand into broader healthcare applications, including:
- Surgical Training : Simulating procedures for surgeons to practice high-risk operations.
- Drug Development : Predicting patient responses to treatments based on simulated clinical trials.
- Preventative Healthcare : Identifying individuals at risk of specific conditions based on behavioral and physiological data.
The company recently secured $7.4 million in seed funding, which will be used for hiring, marketing, and platform development.
“We want people to have the mindset that humans can be tested on when you’re using virtual humans,” says CEO Georgia Witchel, reflecting a bold vision for the future of biomedical experimentation.
Ultimately, Mantis Biotech’s digital twin technology represents a shift toward more proactive, data-driven healthcare. By bridging the gap between real-world limitations and predictive modeling, the company aims to accelerate innovation across the entire biomedical industry.
