A recent study has shown how artificial intelligence (AI) could simplify and speed up the process of detecting important molecular biomarkers in cancer. Traditionally, identifying these biomarkers—key indicators that help predict how a cancer might progress and how it could respond to treatment—requires expensive and time-consuming tests that can also damage tissue samples. These tests often take weeks to deliver results.
Researchers have developed a new AI-based system that could change this. Using the Virchow2 model, an AI trained on 3 million pathology slides, they analyzed over 47,000 whole slide images from nearly 39,000 cancer patients. What makes this approach different is that instead of needing separate models for each biomarker or cancer type, the AI can predict a wide range of biomarkers across various cancers using a single, unified model.
The AI was trained to replicate a biomarker panel used in targeted cancer therapies, successfully identifying 80 biomarkers with high accuracy across 15 common cancer types. It also found strong links between certain biomarkers and specific cancer subtypes, as well as biomarkers that are often tested in clinical settings to help choose the best treatment.
Beyond just identifying biomarkers, the AI could also predict the activity of important signaling pathways and detect issues like DNA repair defects and genomic instability—factors that are important for understanding how a cancer might behave.