UC San Diego Uses 4D Models to Speed up Drug Discovery

Researchers at the University of California San Diego have developed a new method for predicting how drugs affect cells by analyzing the dynamic movement of mitochondria. Instead of relying on static images, these virtual cells capture the real-time shape changes of these energy-producing structures, offering a clearer view of cellular health.

The work, reported by Phys.org, combines artificial intelligence with physics-based simulations to create digital twins of living cells. This approach could accelerate the development of treatments for complex conditions like cancer, diabetes, and Alzheimer’s disease by identifying effective compounds more quickly than traditional laboratory assays.

Capturing dynamic cellular behavior

Mitochondria are often depicted as isolated, static objects, but in reality, they form a constantly shifting network that splits and fuses throughout the cell. Traditional imaging techniques typically provide flat, two-dimensional snapshots that fail to capture this fluid motion. The new study utilizes 4D lattice light-sheet microscopy to record these movements in three dimensions over time, providing a much richer dataset for analysis.

AI predicts health from shape

One approach involved training a deep-learning model called MitoSpace on 40,000 video clips of drug-treated cancer cells. The model learned to identify patterns in mitochondrial shape and movement without human labeling. It successfully grouped cells by their response to different drugs and predicted their energetic state with 75% accuracy, significantly outperforming methods based on flat images.

This capability suggests that mitochondrial form is a reliable indicator of function. The model could serve as a general-purpose tool for sorting cells by developmental stage or identifying new uses for existing medications, reducing the need for extensive manual testing in early-stage research.

Digital twins validate drug effects

The second approach created physics-based digital twins that simulate the behavior of organelles within a cell. By defining rules for how these structures interact, the researchers built a virtual replica that responded to drugs in ways closely matching real biological samples. This validation confirms that the digital models accurately reflect the underlying biology.

Together, these methods offer a faster alternative to time-consuming lab experiments. However, the trade-off is the high cost and complexity of the specialized 4D microscopy required to generate the training data. While this limits immediate accessibility for smaller labs, it sets a new standard for how computational models can be used to understand cellular responses to treatment.