Cancer in motion: how computation is reshaping our view of tumours

Cancer research is undergoing a transformation thanks to computational methods. Maria Secrier, a computational cancer biologist at Cancer Research UK, explains that instead of simply measuring cancer, researchers are now modeling its potential future states. As datasets have grown larger and more complex, the focus has shifted from static snapshots to understanding the dynamic nature of tumours.

Single-cell technologies have revealed that cancer cells do not exist in discrete categories but rather exist along a spectrum of phenotypic states. For example, the epithelial-to-mesenchymal transition (EMT) process, where epithelial cells become more invasive, is not a clear-cut transition but occurs along a continuum of intermediate states that are highly context-dependent and often temporary.

Computational modelling shows that cells rarely fully commit to a single state; instead, they occupy hybrid states that are influenced by their environment.

Spatial data has also played a crucial role in reshaping our understanding of cancer. By integrating spatial omics and imaging techniques, researchers can visualize how cells interact within a tissue and how location influences behavior. In breast tumours, for instance, stromal cells like myofibroblasts can exert long-distance effects on surrounding cancer cells, reshaping entire tissue regions and promoting invasion. The microenvironment can be as informative as the genome itself in predicting cancer cell states.

This ecological view of cancer emphasizes that cell states emerge from interactions within structured environments. Stable states, such as mesenchymal ones, are easier to predict, while hybrid states remain more challenging to capture. Identifying "plastic niches" - tumor regions where cells are more likely to transition between states and develop resistance - could help prevent resistance rather than reacting to it after it occurs.

Treatment is another dynamic aspect of cancer that computational methods are beginning to capture. Cancer treatment does not simply select resistant clones; it can reshape cell states, with cells moving through transient adaptive phases before becoming stable. Understanding these treatment-induced transitions is crucial for predicting relapse and developing more effective therapies.

However, there is still a significant gap in understanding the temporal dynamics of cancer. Most research relies on static snapshots, failing to fully capture the gradual transition from normal tissue to pre-cancerous cells and eventually to malignancy. Longitudinal datasets, which would provide a more complete picture, are currently limited.

Despite these challenges, computational approaches are advancing rapidly. Artificial intelligence, particularly foundation models trained on large-scale biological data, offers new ways to capture complex patterns across scales. These models can recognize subtle regulatory programmes that traditional methods might miss. However, they often struggle with dynamic, plastic states central to cancer progression.

Context-dependent models may perform well in specific scenarios but fail to generalize across different experimental conditions.

The most effective approach may lie in combining methods. Using AI for pattern discovery and large-scale simulations alongside mechanistic, interpretable models could help bridge the gap between complex data and biological insight. Ultimately, the future of cancer research lies in integrating diverse data types and computational techniques to better predict and understand the complex behavior of cancer cells in their natural environment.