Mantis Biotech’s Digital Twin Platform Addresses Biomedical Data Scarcity

Mantis Biotech, a New York‑based startup, announced it is building a platform that creates “digital twins” of human anatomy, physiology and behavior to overcome the persistent data availability problem that hampers biomedical research and drug development.

The company explains that large language models (LLMs) trained on massive textual corpora can accelerate genomics, improve clinical documentation, support real‑time diagnostics, inform clinical decision‑making, speed drug discovery and even generate synthetic data for experimental validation, yet they frequently hit a wall when faced with rare diseases or atypical patient presentations because the underlying training data simply does not contain enough examples.

Mantis’ solution integrates disparate data sources — textbooks, motion‑capture cameras, biometric sensors, training logs, and medical imaging — into a unified repository that feeds an LLM‑based routing, validation and synthesis engine. The resulting synthetic data are then pushed through a physics‑based simulator to produce high‑fidelity renders of anatomical structures and physiological processes.

By grounding the generated synthetic data in realistic physics, the platform aims to reduce the gap between what models have seen and what clinicians encounter in the bedside, enabling more reliable predictions for rare conditions, personalized treatment plans and safer testing of surgical robots before they ever touch a patient.

Georgia Witchel, founder and CEO of Mantis Biotech, told TechCrunch that the physics engine layer is essential because it allows the system to take a known anatomical model, virtually remove a finger or alter a joint, and observe how the surrounding muscles and bones respond, thereby creating labeled data that would otherwise be impossible to collect due to ethical and regulatory constraints.

Witchel added that once such synthetic, physics‑grounded datasets are available, researchers can train LLMs to recognize subtle patterns of injury — like an NFL player’s Achilles‑heel risk — based on performance metrics, training load, diet and activity history, ultimately helping teams make better informed decisions about player health and longevity.

The startup believes that by closing this data gap, its digital‑twin technology could be adopted widely across the biomedical industry — from academic labs studying rare diseases to hospitals implementing AI‑driven decision support and pharmaceutical firms running virtual clinical trials — ultimately accelerating the pace of innovation while keeping patient safety at the forefront.