GenBio AI Virtual Cell Lets Scientists Test Drugs in Silico

Nine separate outlets have written about GenBio AI's AIDO Cell in the past few days, a burst of coverage that trend tracker Archynetys says more than doubled since the story first surfaced. The GenBio AI virtual cell is a software model of a human cell that responds to drugs and genetic edits the way a real cell would, letting researchers run experiment after experiment without ever picking up a pipette. If it works the way its makers claim, a lot of early drug testing could move from the bench to the screen.

GenBio AI was founded in 2024 by Mohamed bin Zayed University of Artificial Intelligence president Eric Xing, Nobel laureate David Baker, and researchers from Stanford, Carnegie Mellon, Harvard, and the Weizmann Institute. It operates out of Palo Alto with a lab in Abu Dhabi, according to UAE Preferred. The pitch is that biology has become computable: instead of testing one drug against one target, scientists can watch a whole cell react. "What we have built is revolutionary because our integrated system will use these state-of-the-art models to create interactive digital versions of biological systems that can be safely experimented on and precisely modified," Le Song, GenBio AI's cofounder and chief technology officer, said in the company's press release, carried by PR Newswire. "This technology lets us program biology the way we program computers, opening up possibilities we've never had before in medicine and biotechnology."

How a virtual cell actually works

AIDO Cell is built around a type of AI called a world model, which keeps an internal picture of the cell that a researcher can alter with a chosen intervention. Add a drug candidate, and the model predicts what happens across layers of biology, then uses that new state as the starting point for the next experiment. From one simulated cell, researchers can read out changes in gene expression, chromatin accessibility, protein abundance, protein location, or the cell's shape, connecting a molecular nudge to its downstream effects. The first release, AIDO Cell 1.0, models two cancer cell lines, K-562 and Hep-G2, chosen because decades of research make them unusually well documented.

What sets the approach apart is memory. Older models tend to make one prediction and start over; this one keeps track of what was already done to the cell, closer to how a real multi-step bench experiment unfolds. "If you knock out a gene and then you treat that knockout cell with a small molecule, you're not treating the original cell," Ziv Bar-Joseph, GenBio AI's cofounder and chief scientific officer, told Drug Discovery News. "You're treating a cell that has already changed."

Why drug hunters care

Drug development is expensive: bringing one drug to market can cost one to two billion dollars, and nine in ten candidates fail. A virtual cell that predicts responses before the lab work starts could trim both numbers. "The goal of the virtual cell, at least the way we see it, is to reduce the number of experiments you're doing," Bar-Joseph told Drug Discovery News. The idea itself goes back decades. In the 1990s, genome scientist J. Craig Venter was already chasing the idea of simulating a whole cell on a computer, and researchers later built a whole-cell model of the bacterium Mycoplasma genitalium, which has only 525 genes. Human cells have tens of thousands of genes wrapped in layers of regulation, which is why scaling that ambition needed AI. GenBio's single-cell models were pretrained on fifty million human cells to learn the landscape first.

There are early signs the approach can generalize. In one experiment described by Drug Discovery News, GenBio excluded the second line from the training data and tested the system with drugs that had only been studied in other cell lines; performance on some tasks improved as more cell lines were added to training. If that pattern holds, well-studied cells could teach the model to make better guesses about cells where data is thin. Scientists are adopting AI quickly across the board, and a Deloitte survey found nearly two-thirds of workers already use it on the job, though the cost of trusting a wrong output is higher in a lab than in a spreadsheet.

What the model can't do yet

The caveats around the GenBio AI virtual cell are real. According to Artificial Science, AIDO Cell has not been peer reviewed, its results come from an in-house benchmark, and the system currently runs only for GenBio's own team and a small group of alpha collaborators. The outlet argues the decisive test is whether the model predicts unseen biology correctly in independent wet-lab experiments, and whether a shared benchmark emerges to judge it. GenBio points at the same constraint from the other side: "The real bottleneck is data," Bar-Joseph said. Expanding beyond the two launch cell lines means new datasets, and the company is courting pharmaceutical partners for proprietary data, which could mean some future virtual cells never become public.

The longer-term ambition goes beyond one cell. "The bigger vision is AIDO, the AI digital organism," Bar-Joseph said, describing a ladder from cells to tissues and organs and eventually a whole simulated organism, which he estimates is at least two or three years away. The announcement lands in a week when researchers also captured a quantum jump of sound for the first time. Whether virtual cells become a standard step in drug discovery or a detour, the test the field is waiting on is the same: predictions that hold up in a real lab.