Recently, Sweden's leading business newspaper Dagens Industri published an in-depth interview with HPA Director, professor Mathias Uhlén, about the global race to build a virtual cell - a next frontier in biomedical AI. The article compares major global initiatives such as AlphaFold, the Chan Zuckerberg Initiative, and China's Pi-HuB project, and highlights the Swedish effort Alpha Cell.

Following the breakthroughs recognized by the 2024 Nobel Prize in Chemistry, awarded to DeepMind's Demis Hassabis and John Jumper for solving the protein folding problem with AlphaFold, and to David Baker for advances in protein design, attention is now shifting to the next big challenge: creating an AI model of a fully functioning human cell.

The Swedish initiative Alpha Cell, coordinated by SciLifeLab and funded with 400 million SEK by the Knut and Alice Wallenberg Foundation, will officially launch early 2026. The project builds on decades of data and knowledge from the Human Protein Atlas, and involves more than 100 researchers.

"After 15 years of building SciLifeLab, it's only natural that Sweden should be part of this race," says Uhlén. "But we are up against the heavyweights."

Unlike language models, which are trained on text, a virtual cell model requires "hard data" - including what proteins exist, where in the cell they are located, and how they are expressed. This is precisely the data foundation Alpha Cell can rely on, thanks to the Human Protein Atlas: It's open access, and it is also used by groups like DeepMind and the Chan Zuckerberg Initiative.

"But of course, we have an advantage from having built the Protein Atlas for 20 years - with 5 million web pages and an enormous amount of data," says Uhlén."

The vision of a virtual cell is to create a digital simulation capable of explaining how diseases develop at the cellular level, and eventually even test drug responses in silico.

"A virtual cell will consist of 20,000 basic components, the proteins, that interact with each other like in a small city," says Uhlén. "Each protein has a specific function and interacts with perhaps ten others. We understand some parts, but far from everything."

Uhlén echoes Demis Hassabis in believing that the first step will be to develop a general consensus model, possibly starting with simpler cells like yeast. However, he expresses skepticism toward the idea of replacing all clinical trials with in silico testing:

"I think it's incredibly naive to think we can run full-scale clinical trials entirely in silico. If we manage to simulate a single cell in five years, that's still far from having the whole body. New molecules can behave unpredictably across all 30 trillions of cells in the body. The current system, with animal and human studies, works well in my view."

Still, in silico methods will likely continue to play an important role in early drug development, helping pharmaceutical companies identify promising compounds faster, safer, and at lower cost.

"And eventually, these virtual cells could be scaled up to form tissues and organs," Uhlén concludes.

Read the whole article (in Swedish) here