Virtual Cells Recreate Mitochondria, Potentially Accelerating Cancer and Diabetes Drug Development

U.S. researchers have developed a “virtual cell” that recreates the movement of mitochondria inside cells in three dimensions, potentially paving the way to accelerate drug development for cancer, diabetes and other diseases.

According to Azure Life Sciences on the 18th (local time), a research team at the University of California, San Diego (UCSD) published two approaches for quantifying changes in mitochondrial shape in the international journal Cell.

Mitochondria are organelles that convert nutrients into energy. They are often depicted as separate, kidney-shaped structures, but in reality they form an interconnected network throughout the cell and rapidly divide and fuse.

The team first trained a deep-learning artificial intelligence (AI) model, MitoSpace, using 40,000 four-dimensional (4D) images obtained after treating cancer cells with 25 compounds.

The model grouped cells with similar responses without being told which drug had been used, and predicted the cells’ energy states based only on the shape and movement of their mitochondria.

The model trained on 4D images classified drugs by mechanism of action with 75% accuracy, higher than the 56% accuracy of conventional models based on 2D images.

Johannes Schönberg, an associate professor at UCSD, said, “Cells are four-dimensional objects with depth that never stop moving,” adding, “Virtual cells must be built from data that captures that fact.”

In a separate study, the team built a “digital twin” of a real cancer cell. They mapped the locations of mitochondria and the microtubules along which they move, and incorporated the motor proteins responsible for transport into a physics-based model.

The team tested whether the model could predict how mitochondria would respond to nocodazole, a drug that partially breaks down microtubules. Without changing a single parameter, the digital twin reproduced the decrease in movement and changes in the rates of division and fusion observed in real cells.

Associate Professor Schönberg said, “Being able to compare a physics-based virtual cell side by side with real 3D microscopy images was previously impossible.”

The team plans to integrate MitoSpace and the digital twin into a single workflow and include other cell organelles in the model to build a complete virtual human cell.

Associate Professor Schönberg said, “The ultimate future is not a single cell, but multiple cells functioning like a tissue,” adding that “modeling an entire tissue could help simulate more realistic human biology and support clinical treatment.”