The UAE Embassy in Washington drew attention on Wednesday to AIDO Cell, a virtual-cell “world model” designed to represent biological activity across multiple levels, from DNA and RNA through proteins and whole-cell behaviour. GenBio AI unveiled a preview of the system in August and says it could eventually help researchers test biological interventions computationally before moving to laboratory experiments.
MBZUAI said its President and University Professor Eric Xing and Professor of Machine Learning Le Song led work on the project as members of GenBio AI’s founding team. Xing is also a co-founder and chief scientist of the California-based biotechnology company, which has operations in Abu Dhabi and Paris. Nobel laureate David Baker is another GenBio AI co-founder.
Unlike models focused on a single biological layer or prediction task, AIDO Cell is intended to maintain a persistent representation of a cell while interventions are applied. GenBio AI says researchers can silence or overexpress a gene, introduce a compound and observe predicted changes propagating across molecular and cellular levels. Because the model retains the altered cellular state, additional interventions can be simulated sequentially rather than beginning again from an unchanged baseline.
The company describes the approach as a general-purpose simulator for cell biology. Its current system integrates multimodal biological data within a shared model, with outputs covering gene expression, RNA splicing, protein structure and cellular morphology. That architecture is designed to address a longstanding difficulty in computational biology: separate specialised models can produce useful predictions without consistently accounting for how changes in one biological layer affect others.
GenBio AI has stressed, however, that the system remains at an early stage. Its published case studies reproduce established biological findings rather than presenting novel predictions already confirmed through laboratory experiments. The company says broader cell coverage and further experimental validation are planned as development continues.
Current demonstrations include simulations involving K-562 and Hep-G2 cell lines, with the model used to examine responses to drug exposure and genetic perturbations. The system can also create branches from a virtual cell state, allowing researchers to compare different experimental interventions from the same starting point.
AIDO Cell builds on earlier work by GenBio AI and academic collaborators on AIDO. Cell, a single-cell foundation model trained on data from about 50 million human cells. That earlier model processes the transcriptome and has been evaluated on tasks including cell classification and perturbation modelling. The new virtual-cell system extends the concept by linking biological scales within one stateful simulation framework.
Researchers behind the underlying single-cell work include scientists affiliated with GenBio AI, MBZUAI, Carnegie Mellon University and the University of Washington. A 2024 preprint describing AIDO. Cell reported model sizes ranging from three million to 650 million parameters and training on diverse human tissues. The authors said the 100-million-parameter version performed strongly on zero-shot clustering, cell-type classification and perturbation tasks, providing part of the technical foundation for later multiscale simulation efforts today.
Xing has framed the wider objective as making biology more computable and predictable by modelling processes across multiple scales. GenBio AI’s longer-term AIDO programme seeks to connect foundation models for DNA, RNA, proteins, cells and other biological systems, though the company acknowledges that comprehensive simulation of cell biology remains unsolved.
The company is granting access to AIDO Cell on a controlled basis to its own teams and selected alpha collaborators while preparing an early-access programme for academic, biotechnology and pharmaceutical researchers. It says access conditions prohibit harmful applications and that the current version does not accept viral or bacterial inputs.
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
