Scientists Build Virtual Cells With AI to Predict Disease and Drug Response

Scientists are building “virtual cells” — artificial-intelligence models that simulate the inner workings of living cells — in a bid to predict how cells behave, how diseases evolve and how new drugs will perform before anyone touches a laboratory dish. The push is drawing billions of dollars, thousands of competing research teams and a growing debate about how much the models can really deliver.

The push rides on a class of artificial intelligence that has already transformed biology.

Models trained on hundreds of thousands of protein structures can now look at a gene and predict the intricate structure of the protein it encodes. That achievement earned David Baker, a molecular biologist at the University of Washington, and two other scientists a share of the 2024 Nobel Prize.

Baker and his colleagues are now using AI to design new proteins that can do everything from break down pollutants to act as vaccines, according to reporting first published by The New York Times.

Can AI predict life? Inside the federal plan to map human cells — The Hill

How a Virtual Cell Works

Each cell in the body is a living world. Its surface is crammed with sensors and channels, enclosing a stew of billions of molecules that continually reconstruct themselves in thousands of forms, encoded by genes on long ribbons of DNA in the nucleus. Each cell type switches on a different combination of genes to carry out its particular tasks — sprouting hair, battling viruses, generating a dream.

Virtual cell developers train their models much the way engineers build world models for other complex systems. The AI is taught to create a detailed internal model of a cell’s biology, then to trace how a given cell changes from one state to another, plotting its evolution along various possible paths. Researchers at GenBio.AI are developing a world model for cells along those lines.

Baker said virtual cells could prove valuable for testing some of the most sophisticated new forms of medicine. He and his colleagues are engineering immune cells that can recognize and destroy cancer cells. With an accurate virtual cell, they could quickly test whether their designs make the immune cells work as planned.

Dollars Flow as Skeptics Push Back

Competition is stoking progress as well. More than 1,200 teams from around the world entered the first Virtual Cell Challenge; this year, more than 1,600 teams tried their luck. The Institute was set to announce the winners on Thursday.

The inaugural challenge, launched by the Arc Institute with a $100,000 grand prize and sponsored by NVIDIA, 10x Genomics and Ultima Genomics, asked entrants to build models predicting how cells respond to single genetic perturbations. The work drew on a new dataset of 300,000 human embryonic stem cells carrying 300 genetic perturbations.

The scientific literature is converging on the same territory. A review published Thursday in Expert Opinion on Drug Discovery by researchers at the University of Melbourne and Helmholtz Munich lays out a framework of six capabilities for AI-powered virtual cells, spanning population-derived models, digital twins and agentic twins.

The authors argue the models could serve as foundation models that transfer biological knowledge across tasks — including settings where data is limited. Their value, the authors caution, depends on faithful extrapolation beyond training data and validation against real biological observations.

A separate October review in Advanced Healthcare Materials points to virtual-cell-assisted high-throughput screening and toxicity prediction among recent advances. It also flags hurdles in scalable manufacturing, long-term stability and regulatory adaptation.

Skeptics remain. Dr. Anders, a virtual cell researcher quoted in the Times report, warned that it is easy to oversell the technology to investors who cannot judge the claims.

Baker, who was born in Seattle in 1962, directs the University of Washington’s Institute for Protein Design and is an investigator with the Howard Hughes Medical Institute. His group created the Rosetta algorithm and RoseTTAFold, and designed Top7, the first artificial protein with a novel fold.

He has co-founded more than a dozen biotechnology companies, including GenBio AI, whose researchers are now developing a world model for cells, and received the Breakthrough Prize in Life Sciences in 2021 before sharing the 2024 Nobel.

Even the field’s boosters do not expect a perfect simulation of a human cell any time soon. Goodarzi, an Arc Institute core investigator, believes virtual cells will yield valuable insights well before then.

Photo of author

Health Desk

Health Desk at Archyde.

Zurich Unveils Giant Walk-in Advent Wreath at New Bellevue Noël Christmas Market