Arc Institute Competition Showcases Advances Toward ‘Virtual Cell’

Researchers striving to create a ‘virtual cell’—a computer model capable of simulating biological processes with the precision of the real thing—reached a new milestone this month. The Arc Institute, an AI and biology research center based in Palo Alto, California, announced the top finishers of its second annual Virtual Cell Challenge on Nov. 19.
The competition, which invites researchers to design computer models that predict cellular responses to genetic alterations, saw a substantial increase in engagement this year. According to contest head Hani Goodarzi, an AI research lead at the Arc Institute and a professor of biochemistry at the University of California, San Francisco, approximately 1,000 teams participated in the final judging, representing a threefold increase over the previous year. “It’s a staggering number of participants,” Goodarzi said. “This year, everyone jumped in.”
Scientific interest in virtual modeling surged after 2020, when Google DeepMind’s AlphaFold system successfully predicted the folded structure of proteins. While AlphaFold utilized the extensive Protein Data Bank, the creation of a virtual cell poses a considerably more complex challenge. “Cells have billions of molecules inside of them, and thousands of different processes that are all coordinated and taking place at any one time,” said Shana Kelley, a professor of chemistry and biomedical engineering at Northwestern University and president of bioengineering at the Chan Zuckerberg Initiative’s Biohub. Kelley, who was not involved in the competition, noted that training such a model would likely require billions or even trillions of datasets.
For the Arc Institute competition, organizers focused on a specific goal: predicting how different cell types respond when genes are switched off. This objective aims to replicate expensive, time-consuming lab experiments in an in silico environment, potentially streamlining future drug discovery. “If you think about, similar to AlphaFold, replacing something that’s very slow and expensive with a reliable in silico simulation—that is like the north star for us, for at least this phase,” Goodarzi stated.
The competition’s first-prize award of $100,000 was claimed by a team from the biotechnology company Illumina, which leveraged the firm’s own Billion Cell Atlas dataset. Third prize, worth $25,000, went to a team from Xaira Therapeutics, which also used proprietary data. The second-prize model, named FlowState, was developed by independent researchers Sina Khodaveisi and Navid Bagheri, who received $50,000. Khodaveisi, a machine learning expert, credited the model’s success to its explainable structure: “One decides what changes in the cell, another decides how big the impact is, and the last decides the direction of the change. This is what makes our model interesting—we could see the reasoning behind it.”
Despite the progress, developers acknowledge that models still struggle to capture the unique quirks of specific cell types. Addressing the scarcity of data remains a primary hurdle. “Biological data has to be made. It’s not found,” said Illumina team member Eyal Ben David. Several initiatives are currently working to expand the available data pool. In April, Biohub announced a five-year Virtual Biology Initiative, and in July, the U.S. National Institutes of Health launched the Bio Genesis Mission. “In the next 18 months, we think we can increase the amount of available relevant data by about tenfold as a start,” said Kelley. “Then we’ll keep going from there.”
Goodarzi confirmed that the Arc Institute plans to sustain the momentum, noting that the Virtual Cell Challenge will become a recurring event. “We are going to run this indefinitely, every year,” he said. “Until we have a virtual cell.”