The core problem in oncology has always been one of discrimination. Cancer cells and normal cells are, at the molecular level, nearly identical. What distinguishes a cancer cell is dysregulation, a set of genetic switches flipped in the wrong direction, causing uncontrolled growth. For decades, finding and exploiting those switches required hunting through patient samples by hand, looking for patterns subtle enough to be almost invisible. 

AI has changed what’s possible. Systems trained on genomic databases spanning tens of thousands of sequenced cancer samples can now identify the master regulatory patterns that are active specifically in cancer cells and not in surrounding healthy tissue. Unlike the biomarkers of older precision oncology, these are fine-grained genomic signatures that encode the difference between malignant and normal at the level of how genes are switched on and off. 

Once those signatures are identified, they unlock a range of approaches that simply weren’t possible before. AI is helping researchers design personalized cancer vaccines that train the immune system against the unique mutations a patient’s tumor produces. 

Moderna and Merck are already in late-stage trials doing this, building on the same mRNA infrastructure that powered the COVID-19 vaccines. AI is also helping engineers build smarter CAR T cells that use tumor-specific signals to stay active inside the immunosuppressive environment of a cancer, rather than exhausting themselves before the job is done. At the earliest end of the pipeline, AI-driven analysis of genomic and imaging data is making it possible to detect cancers years before........

© Time