Tackling the complexity of cancer with generative models

The Hallmarks of Cancer framework has played a seminal role in developing our understanding of cancer biology. By design, these hallmarks abstract cancer into a common set of functional capabilities. The hallmarks thus constitute an intentionally reductionist framework that has unified diverse observations and yielded valuable mechanistic insight, while leaving unresolved how these processes interact across scales. Complementary tools are therefore needed to capture cancer's inherently complex, multimodal, and multiscale nature. Here, we posit that generative models, built on the recent advances of artificial intelligence, are the key technology to capture this complexity and to thereby improve how we diagnose, understand, and intervene in cancer. Specifically, because of their ability to recognize complex patterns, process unstructured inputs, and synthesize multimodal inputs, generative models are poised to usher in a new era of biological discovery and clinical care. Ultimately, we envision a synergistic cycle wherein generative models of cancer and the Hallmarks of Cancer complement one another, the former driving hypothesis generation and discovery and the latter guiding the prioritization and development of new measurement tools.