On a supercomputer in Illinois, a cell is being born. Not literally (no membranes are manufactured, no DNA is physically copied), but inside a cluster of NVIDIA A100 graphics cards, every molecule that constitutes a living bacterium is accounted for, tracked through space, made to obey the laws of chemistry and physics over a full 105-minute life cycle. It grows. It replicates its chromosome. It pinches in two. Then it does it again. And again. Fifty times, each replicate slightly different from the last, each one a unique individual.
This is not a cartoon of a cell. Zan Luthey-Schulten, a chemistry professor at the University of Illinois Urbana-Champaign who has spent years building toward this moment, is careful on the point. “This is a three-dimensional, fully dynamic kinetic model of a living minimal cell that mimics what goes on in the actual cell,” she says. Every one of the 493 genes in the simulated bacterium is expressed. Every ribosome assembles, diffuses, translates. Every strand of DNA threads through the crowded cytoplasm, copies itself, and segregates to opposite ends of the dividing cell. The work, published today in Cell, represents the first time a complete cell cycle has been simulated at this level of spatial and molecular detail.
The organism at the centre of it all is called JCVI-syn3A, a synthetic bacterium developed at the J. Craig Venter Institute in California. Syn3A, as researchers call it, is about as stripped-down as life gets: a single circular chromosome, roughly 543,000 base pairs long, carrying just the genes needed to replicate DNA, metabolise nutrients, build proteins and divide. No flagella, no complex regulatory networks, no redundancy to confound the model. What makes it ideal for this kind of work isn’t what it can do; it’s how little it does. Each of its roughly 500 genes matters. The machine had to account for all of them.
Building that machine took years.
Postdoctoral fellow Zane Thornburg and graduate student Andrew Maytin spent much of that time solving problems that had no established solutions. The simulation works by coupling several distinct computational methods: molecular dynamics for the chromosome, reaction-diffusion equations for the movement of proteins and RNA through space, ordinary differential equations for metabolism, all communicating with each other every fraction of a second of simulated time. Periodically the chromosome’s position updates the spatial model; periodically the metabolic state feeds back into whether DNA replication can proceed. The organism is not treated as a well-stirred pot of chemistry. It is treated, as actual cells are, as a place where things happen at specific locations, and where location matters.
Getting the chromosome to move was particularly brutal. Syn3A’s single circular chromosome must, during each 105-minute cycle, replicate itself entirely and then segregate the two daughter copies to opposite poles of the dividing cell. This is hard enough to do in biology; it turned out to be even harder to simulate. The chromosome required its own dedicated graphics processing unit, running in parallel with everything else. “One of the last big hurdles that Andrew and I had to solve,” Thornburg says, “was understanding how the membrane and the DNA talk to one another when both are moving.” That sentence understates the problem considerably. Thornburg put it more directly elsewhere: “I can’t overstate how hard it is to simulate things that are moving … and doing it in 3D for an entire cell was triumphant.”
The question that any model eventually has to answer is whether it’s right. Here, the team had an unusual advantage: Syn3A has been studied so intensively by experimentalists that the simulations could be checked against a remarkable range of real data. Luthey-Schulten’s collaborators at Harvard Medical School and Boston Children’s Hospital, led by biophysicist Taekjip Ha and chemist Angad Mehta, generated new imaging and DNA sequencing data specifically to test and refine the model. Their work confirmed that Syn3A’s cell division is symmetrical, and pinned down the extent of DNA replication during each cycle. Both turned out to be critical constraints. In 50 replicate simulations of the full cell cycle, the average predicted doubling time came within roughly two minutes of the experimentally measured 105-minute figure. For a system modelling thousands of molecular species interacting across three-dimensional space, that is, in a word, remarkably close.
The simulation isn’t perfect. The model currently omits polysomes (clusters of ribosomes translating the same messenger RNA simultaneously), which means some proteins, particularly large ones requiring sustained translation, are slightly underproduced relative to what experiments show. The chromosome segregation mechanism relies in part on an artificial repulsive force applied to daughter chromosomes, because the biological mechanisms that partition chromosomes in this organism remain genuinely unknown. No MinD system, no ParABS. Perhaps the minimal cell solves this problem through entropic forces alone; perhaps there is something undiscovered. The simulation has to work around the gap rather than model it.
What the model can do, though, is something no single experiment can match. “We have a whole-cell model that predicts many cellular properties simultaneously,” Luthey-Schulten says. “If you want to know what’s going on, say, in nucleotide metabolism, you can also look at what’s going on in DNA replication and the biogenesis of ribosomes. So the simulations can give you the results of hundreds of experiments simultaneously.” The model doesn’t just predict average behaviour, either. Because each of the 50 simulated cells starts from slightly different initial conditions, and because molecular events are treated stochastically, each daughter cell ends up with a slightly different complement of proteins, RNAs and ribosomes. The partitioning of molecules at division, the team found, approximates a binomial distribution, roughly what you’d expect from random diffusion, with no detectable bias toward either daughter. Cells are, from this angle, fundamentally noisy things.
At any given moment during a simulated cell cycle, the chromosome is coiled and moving inside a sphere barely 200 nanometres across; about 500 ribosomes are diffusing through a cytoplasm so crowded that some components had to be rendered invisible to visualise others at all. Around 55% of those ribosomes are actively translating at any moment. Roughly 70% of RNA polymerases are occupied. About a hundred ribosomal subunits sit in various states of incomplete assembly, waiting for the right protein to arrive and finish them off. Every mRNA has a half-life. Every gene has a probability of going untranscribed in a given cycle. The simulation tracks all of it.
Each of those 50 cell cycles required four to six days of compute time on two dedicated GPUs, about 15,000 GPU-hours in total just for the main analysis. The simulation is not cheap to run. But Luthey-Schulten is clear that the point was never to replace experiment. It was to do something experiments cannot: reveal the cell as an integrated whole, where metabolism, gene expression, chromosome dynamics and membrane growth are not separate phenomena but continuous, mutually constraining features of the same living process. “Such a comprehensive undertaking,” she says, “was only possible through the combined efforts of a host of collaborators.”
The next step, probably, is complexity. Syn3A is the floor: the minimum viable cell. Every other organism is something built on top of it, with more genes, more feedback loops, more ways to fail and adapt. The 4D model provides what its creators call a platform: a fully characterised foundation onto which additional biology can, in principle, be layered. More realistic chromosome mechanics. Polysomes. A kinetic model of the division ring. Coupled transcription and translation. Whether a full human cell cycle is ever simulatable this way is an open question, and not the most urgent one. For now, it is enough that we can watch, molecule by molecule, what it actually takes for the simplest known self-replicating organism to do what it has always done; and to ask, with some precision, what we still do not understand about it.
Source: “Bringing the genetically minimal cell to life on a computer in 4D,” Cell (2026). DOI: 10.1016/j.cell.2026.02.009
Frequently Asked Questions
Why use such a stripped-down bacterium rather than a more complex cell? The minimal cell JCVI-syn3A has just 493 genes, all of which are essential for life; there is no redundancy, no mystery pathways, no genetic dead weight to confuse the model. Every gene had to be accounted for, which made the simulation harder to build but much easier to validate against experimental data. More complex organisms have so many interacting variables that checking whether a simulation is actually correct becomes nearly impossible.
How close did the simulation actually get to the real thing? Across 50 replicate cell cycles, the predicted doubling time averaged within about two minutes of the experimentally measured 105-minute figure. The model also correctly predicted the ratio of DNA at the replication origin versus the terminus, ribosome counts, and the approximate distribution of proteins across daughter cells. It’s not atom-by-atom perfect, but for a system tracking thousands of molecular species across three-dimensional space, the agreement with real biology is striking.
Could this kind of simulation eventually be used for drug discovery? That’s the direction the field is moving, though current models are still built on simpler organisms. A complete whole-cell model lets you ask what happens to metabolism when you block a specific enzyme, or how gene expression changes when a protein is mutated, questions that would otherwise require many separate experiments. The minimal cell simulation is a proof of concept; applying the same approach to medically relevant cells is a much harder problem, but the computational methods developed here are designed to scale.
Why does it take so long to simulate just 105 minutes of biology? The simulation resolves molecular positions to 10 nanometres across a three-dimensional grid, tracks every protein and RNA molecule as an individual particle, and updates thousands of coupled chemical reactions every fraction of a simulated second. The chromosome dynamics alone required a dedicated graphics card running in parallel. Each of the 50 cell cycles took four to six days of supercomputer time, roughly 15,000 GPU-hours total. The biological timescale is short; the physical modelling required to capture it faithfully is extraordinarily expensive.
ScienceBlog.com has no paywalls, no sponsored content, and no agenda beyond getting the science right. Every story here is written to inform, not to impress an advertiser or push a point of view.
Good science journalism takes time — reading the papers, checking the claims, finding researchers who can put findings in context. We do that work because we think it matters.
If you find this site useful, consider supporting it with a donation. Even a few dollars a month helps keep the coverage independent and free for everyone.