Are you looking for a better way to biopsy for cancer, especially in deep tissue areas such as the lung and brain that aren’t readily accessible? In a study funded by NCI, researchers used machine learning to find a potential new biomarker that may someday help.

The researchers identified a panel of protein signatures underlying exosomes—small “cargo-carrying” cell structures known to have a part in diseases such as cancer.

Through a machine learning approach, the researchers were able to differentiate cancer exosomes from non-cancer exosomes. They found five exosome proteins that reliably identified five common cancer types from blood or plasma samples. They discovered similar biomarkers for detecting cancer from urine samples.

According to the authors, by characterizing these exosome proteins they were able to identify reliable biomarkers for detecting cancer at its earliest stages.

Corresponding author, Dr. Raghu Kalluri of MD Anderson Cancer Center, Rice University, and Baylor College of Medicine, noted, “This approach not only means we can detect cancer sooner, and in non-invasive liquid biopsy sources, but it worked across a wide range of cancer types, even identifying tumors of undetermined origins.”