Transforming Biomedical Research: Mantis Biotech’s Digital Twins

Large language models trained on extensive datasets hold the potential to revolutionize genomics research, enhance clinical documentation, improve real-time diagnostics, aid clinical decision-making, fast-track drug discovery, and even create synthetic data for experimental advancements.

The Challenge: Limitations in Edge Cases

Despite their promise, large language models often hit a bottleneck in biomedical research. These models struggle with edge cases, such as rare diseases and atypical conditions, where reliable and representative data is scarce.

Mantis Biotech: Bridging the Data Gap

Based in New York, Mantis Biotech is developing innovative solutions to address this data availability challenge. Their platform integrates diverse data sources to create synthetic datasets, enabling the development of “digital twins” of the human body—predictive models that simulate anatomy, physiology, and behavior.

Applications of Digital Twins in Healthcare

Mantis is promoting these digital twins for data aggregation and analysis, suggesting they could be invaluable for studying and testing new medical procedures, training surgical robots, and predicting medical issues or behavioral patterns. For instance, a sports team might predict the likelihood of an NFL player suffering an Achilles injury based on various factors, as explained by Mantis’ founder and CEO, Georgia Witchel, in a recent TechCrunch interview.

How the Technology Works

To construct these digital twins, Mantis’ platform synthesizes data from multiple sources, including textbooks, motion capture cameras, biometric sensors, training logs, and medical imaging. It employs an LLM-based system to validate and synthesize these data streams and utilizes a physics engine to create accurate high-fidelity models, which can be used for training predictive algorithms.

The Importance of the Physics Engine

According to Witchel, the physics engine is essential because it enhances the information by realistically modeling the physics of anatomy, grounding the generated synthetic data in real-world principles.

Generating Data for Edge Cases

Witchel illustrated the technology’s potential by discussing hand-pose estimation for individuals missing fingers. “We could easily generate a dataset for that by removing a finger in our physics model and regenerating it,” she noted.

Broadening Biomedical Applications

Witchel believes Mantis’ platform can be widely utilized across the biomedical industry, particularly in areas where data about procedures or patients is unstructured or siloed. It has significant implications for edge cases and rare diseases, where ethical and regulatory constraints hamper data access.

A Vision for Digital Twins

“I want people to approach our digital twins with the same curiosity as a child playing with a toy,” Witchel stated. “This mindset will encourage the exploration of testing humans using virtual models while respecting data privacy.”

Success in Professional Sports

Mantis has found success within the professional sports arena, including partnerships with an NBA team focusing on modeling high-performing athletes. Witchel explained, “We create digital representations that track an athlete’s jump performance over time, correlating it with their sleep patterns and training intensity.”

Recent Funding and Future Directions

Recently, Mantis raised $7.4 million in seed funding led by Decibel VC, alongside participation from Y Combinator, angel investors, and Liquid 2. This funding will support hiring, marketing, and go-to-market strategies.

Looking Ahead: Preventative Healthcare

Witchel indicated that the company’s next steps involve advancing their technology and eventually making the platform accessible to the broader public, with a focus on preventative healthcare. Mantis is also collaborating with pharmaceutical labs and researchers conducting FDA trials to provide insights into patient responses to treatments.