Early in 2026, a computer operated a living cell in a lab at the University of Illinois Urbana-Champaign. Not a cell simulation in the loose, impressionistic sense that the term is typically used. It’s not a cartoon model that mimics some important functions. A complete, molecule-by-molecule replication of every gene turning on, every protein putting together, every piece of chemical machinery running through its cycle—from start to finish, DNA copying, cell swelling, and ultimately splitting in two. the entire thing. utilizing GPUs. The simulation of roughly 105 minutes of real cellular life took six days.
Just looking at that ratio is worthwhile. Compared to the cell it was copying, the computer was operating about 80 times more slowly. Furthermore, neither a complex bacterial cell nor a human cell were present. JCVI-Syn3A, a simplified synthetic bacterium with just 493 genes, was the most basic self-replicating organism that scientists have ever discovered or created. About 20,000 are carried by a human cell. It doesn’t scale linearly, whatever that gap suggests about how hard it is to simulate us. It grows at an exponential rate.
| Lead researcher | Prof. Zan Luthey-Schulten, University of Illinois Urbana-Champaign (UIUC) |
| Institution | University of Illinois Urbana-Champaign — Beckman Institute for Advanced Science and Technology |
| Published in | Cell journal, March 2026 |
| Cell simulated | JCVI-Syn3A — a synthetic minimal bacterium developed by the J. Craig Venter Institute (JCVI), California |
| Genome size | 493 genes on a single circular strand of DNA |
| Simulation scope | Full 105-minute cell cycle: DNA replication, protein synthesis, metabolism, membrane growth, and cell division — tracked in 4D at nanoscale resolution |
| Computing cost | 4–6 days of continuous GPU computing per run; 15,000 GPU hours to simulate 50 cells |
| Speed ratio | Computer ran ~80× slower than the actual living cell |
| Project began | Foundational modeling work started in 2018 in collaboration with Harvard Medical School |
| Key collaborators | Angad Mehta (UIUC), Taekjip Ha (Boston Children’s Hospital / Harvard Medical School), Zane Thornburg, Andrew Maytin |
| Broader context | Part of a global push including the Chan Zuckerberg Initiative, Arc Institute, and Stanford’s virtual cell initiative |
| Reference / source | Illinois News Bureau — UIUC official announcement |
Chemistry professor Zan Luthey-Schulten spearheaded the work, which was published in the journal Cell in March 2026. Since at least 2018, he has been establishing the conceptual framework through a number of joint publications with Harvard Medical School.
The fact that someone completed the entire task at once, rather than just the technical accomplishment, is what makes this historic. Snapshots and fragments of cellular behavior were captured by earlier models. The film is run by this one. Every molecule was monitored over time in three dimensions, interacting with every other molecule in the cell in real time. It’s the 4D in “4D simulation” that researchers frequently bring up, and it’s more important than it may seem.
The group purposefully selected Syn3A. The J. Craig Venter Institute in California invented the minimal cell, which is essentially a bacterium with only the 500 genes needed for division, metabolism, and replication after all genetic redundancy was eliminated. Imagine it as life without the ornamentation.
It is currently used as a research organism in more than 50 laboratories worldwide, in part because it is tractable and in part because everything you learn about it touches on the most fundamental aspects of what cells actually do. Even with just 493 genes to account for, the computational demands were astounding: the team required two GPUs running in parallel, one of which was solely used to simulate DNA replication; this process alone doubled the runtime when handled on a single machine.
The fact that approximately 92 of Syn3A’s 493 genes still have unknown functions is difficult to ignore. This implies that even though the simulation is amazing, it relies in part on educated guesses. This is known to the researchers. It’s one of those minor acknowledgements in the scientific literature that the media usually ignores, but in reality, it’s an indication of good faith. Although the model is thorough, it is not finished. As the excitement grows, it’s important to remember that there is a difference.
The pharmaceutical perspective is what makes this relevant outside of computational biology labs. Approximately 90% of medications that go into clinical trials fail, either because they don’t function as intended or because they cause unanticipated side effects on cellular machinery that wasn’t sufficiently modeled during the preclinical stage.
That calculus is altered by a virtual cell. Before a single compound comes into contact with a living thing, you could theoretically run thousands of drug candidates through a digital cell. That is the specific next step that scientists in this field have been pursuing for years; it is not conjecture. By cataloguing data from nearly 100 million cells through its open platform and collaborating with NVIDIA to scale that toward billions of observations, the Chan Zuckerberg Initiative, the philanthropic arm of Mark Zuckerberg and Priscilla Chan, has effectively made virtual cells its primary scientific mission.
The UIUC simulation is not a stand-alone entity. A global sprint is underway, and it has been quietly picking up speed for a few years. In 2025, the Arc Institute held its inaugural Virtual Cell Challenge, which attracted more than 5,000 participants from 114 nations. Can an AI model accurately predict a cell’s response when a particular gene is silenced to the point where it can take the place of an actual laboratory experiment? In all honesty, pure AI models were not consistently outperforming statistical baselines. That’s a helpful fact. It implies that the field is not yet at a stage where pattern recognition by itself can bear the burden of biological prediction; rather, it is still in the difficult, ground-level work phase.
As this develops, it seems as though biology is experiencing its own AlphaFold moment, when computation finally caught up to a problem that had appeared to be computationally unsolvable for decades. The virtual cell is attempting to crack something bigger: the behavior of life as an integrated system, just as AlphaFold cracked protein folding. The experimental datasets, GPU clusters, and validated simulation frameworks that are currently being developed may serve as the basis for something that we are still unable to fully predict.
A human cell and a 493-gene bacterium still differ greatly. In retrospect, the current accomplishment seems like the first sentence of a very long book due to the additional layer of complexity brought about by tissue-level biology, where millions of cells coordinate, communicate, and organize into organs. Simulating something as complex as a single human neuron in its entire biological context—let alone the network of 86 billion neurons it is a part of—will take years, if not decades.
However, the evidence is now available. A team used a computer to create a functional model of a living cell that behaved like the real thing from conception to division. That is a significant accomplishment. It may be the most important computational biology outcome of this decade, which is noteworthy considering the decade.
