Mantis Biotech Debuts Human Digital Twins to Solve Data Scarcity

Mantis Biotech announced an expansion of its digital twin platform alongside a $7.4 million seed round led by Decibel Partners. The round included participation from StoryHouse Ventures, Y Combinator, Pioneer Fund, Spot VC, and Fenwick. The system synthesizes fragmented medical data into physics-based predictive models of human anatomy and physiology. By generating high-fidelity synthetic datasets, the platform addresses the chronic lack of training data for rare diseases, edge cases, and highly specific patient populations.

Architecture and Data Synthesis

The core technical challenge in medical modeling is unstructured and fragmented data. Mantis aggregates inputs from electronic health records, genomic sequences, wearable biometrics, medical imaging, and peer-reviewed journals. The pipeline relies on an LLM-based system to route, validate, and refine these disparate streams into a unified framework.

Handling unstructured clinical text requires precise data extraction. Using language models to enforce structured output allows the system to normalize messy inputs from motion capture cameras and workout logs. Developers building sensitive data applications often face strict privacy constraints when sharing raw patient datasets. Mantis bypasses these bottlenecks by generating entirely synthetic physiological profiles rather than attempting to anonymize existing clinical records.

Physics Engine Integration

The defining feature of the platform is its proprietary physics engine. Standard language models fail to accurately predict physical limitations in biological systems because they operate entirely on statistical text patterns. Mantis grounds its synthesized data in realistic anatomical and physiological constraints.

This architecture allows the system to simulate movements and biological responses that do not exist in the training data. The engine can predict the specific hand mechanics of individuals missing digits by calculating the necessary physical compensations. Combining deterministic physics engines with probabilistic LLM outputs provides a highly effective blueprint for healthcare robotics research.

Current Production Workloads

The platform initially proved its viability in professional sports. An undisclosed NBA team uses the system to create digital twins of high-performance athletes. By tracking continuous metrics like jump height, sleep duration, and training load, the models predict the probability of severe injuries, such as Achilles tendon ruptures, before physical failure occurs.

The new funding scales this technology into pharmaceutical research and clinical operations. In March 2026, Mantis deployed virtual patient populations in a major clinical trial. The synthetic cohorts allowed researchers to simulate treatment responses without recruiting massive human sample sizes. Simultaneously, manufacturers are utilizing these high-resolution environments to train robotic surgical systems. The digital twins provide physics-accurate virtual tissues and organs, enabling the rigorous testing of automated procedures without risking human subjects.

When designing predictive models for constrained physical systems, deterministic guardrails matter more than raw parameter count. If your application predicts physical outcomes, evaluate how a dedicated physics engine can constrain your language model outputs to strictly possible realities.