Abhiru
7/17/26
For decades cancer treatment has relied on three primary approaches: surgery, chemotherapy, and radiation. In recent years, however, a fourth option known as immunotherapy has transformed oncology by harnessing the body’s immune system to attack cancer. While immunotherapy has produced remarkable results it has one major limitation: most patients do not respond to it. Doctors have struggled to determine beforehand who will benefit from these treatments.
A recent study funded by the European Union introduces COMPASS, an AI model designed to solve this problem. COMPASS analyzes the genetic activity of a patient’s tumor to predict whether that individual is likely to respond to immunotherapy. Unlike previous prediction methods that only worked for specific cancers, COMPASS was designed to make accurate predictions across many different cancer types and treatments.
Immunotherapy works by helping the immune system recognize and destroy cancer cells. Normally cancer cells can hide from immune cells by activating molecular checkpoints that suppress the body’s response. Drugs known as immune checkpoint inhibitors block these checkpoints allowing immune cells to attack the tumor. However since every patient’s tumor is biologically different some patients experience dramatic improvements while others receive little or no benefit. Current biomarkers such as PD-L1 expression and TMB provide only limited predictive accuracy.
COMPASS approaches this challenge differently. Rather than relying on a handful of biomarkers it analyzes the activity of thousands of genes within a tumor and condenses that information into 44 biologically meaningful immune concepts including immune cell populations, signaling pathways, and interactions between tumors and their surrounding environment. By learning these broader biological patterns the AI can make predictions that extend across multiple cancer types instead of being limited to just one.
To build the model, researchers first trained COMPASS on genetic data from 10,184 tumors representing 33 different cancer types. They then tested it on 1,133 patients from 16 different clinical studies, covering seven cancers and multiple immunotherapy drugs. The results were impressive: COMPASS was able to outperform 22 existing prediction methods and improve prediction accuracy drastically.
The researchers also discovered that COMPASS could provide valuable insight into why certain patients fail to respond to immunotherapy. Some tumors appear to have an active immune response but still resist treatment. COMPASS identified biological mechanisms associated with this resistance, including excessive TGF-β signaling, poor blood vessel access that prevents immune cells from entering the tumor, dysfunctional CD4 T cells, and reduced B-cell activity. Understanding these mechanisms could help scientists develop new therapies that overcome resistance to immunotherapy.
Although COMPASS is still a research tool and requires further validation before routine clinical use, its results highlight the growing role of artificial intelligence in precision medicine. Rather than replacing physicians, AI systems like COMPASS can help doctors make more informed treatment decisions by identifying which patients are most likely to benefit from specific therapies. As cancer treatments become increasingly personalized, tools that combine advanced machine learning with biological understanding may improve patient outcomes while reducing unnecessary treatments.
COMPASS represents an important step toward the future of oncology: one in which treatment decisions are guided not only by the type of cancer a patient has, but also by the unique biological characteristics of their individual tumor.