Scientists publish blueprint for AI ‘digital organism’

A team of researchers led by Eric Xing published a paper in Nature Medicine on Thursday laying out a roadmap for building an AI-driven "digital organism" — a system of connected AI models that could simulate how drugs or genetic changes move through molecules, cells, and eventually entire people.axios+1

The paper, titled "How to build an AI-driven digital organism," describes what the authors call AIDO: a framework that would link specialized AI models for DNA, proteins, cells, tissues, and health outcomes into a single dynamic simulator. Xing, who is co-founder and chief scientist of GenBio AI, president of the Mohamed bin Zayed University of AI in Abu Dhabi, and a professor of computer science at Carnegie Mellon University, compared the concept to "a flight simulator for medicine".nature+2

From Prediction to Simulation

Most existing AI tools in biology handle one task at a time, such as predicting the three-dimensional shape of a protein. The Nature Medicine paper argues the field needs to move beyond these siloed models toward an integrated system capable of simulating chains of biological events.axios+1

"It is really about building a simulator of biology," Xing told Axios. He said GenBio AI will soon release what he called "the world's first implementation of the virtual cell." The company announced a collaboration with Nvidia ↗ 0.54% in June to accelerate development of its virtual-cell world models.genbio+1

A Crowded Field

GenBio AI is not alone in pursuing virtual cells. The Chan Zuckerberg BioHub has been building an open platform of AI models to understand and predict cell behavior, with the stated goal of modeling cells across tissues, people, and conditions. A June Nature feature described the broader push by multiple research groups to convert molecular data into predictive simulations of cellular behavior.nature+1

Xing described Zuckerberg's BioHub as a "friendly competitor," noting that its approach leans more heavily on language-model-style sequence models, while GenBio combines sequence, structure, imaging, and interaction data in one multi-scale system.axios

Limits of the Vision

The researchers acknowledge they are not close to building a virtual human. The near-term aim is a virtual cell that can take multiple kinds of biological data into account and predict the effects of interventions — not to replace animal studies or clinical trials, but to serve as a filter that eliminates weaker drug candidates before researchers spend years testing them. Challenges remain around biased training data, patient variability, and AI's tendency to confuse correlation with causation.axios