The test was developed by a team at Mass General Brigham, particularly in the Mike Toth Head and Neck Cancer Research Center (Mass Eye and Ear). Their earlier work had shown that using only small portions of the HPV genome or more limited assays yielded reduced accuracy, especially for early or minimal disease. By contrast, HPV-DeepSeek’s “look broadly” approach improves detection, especially in early stages.
In follow-up validation, the researchers are applying the test to larger, blinded sample sets, including those from national repositories such as the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO).
The key principle is liquid biopsy – detecting fragments of tumor DNA circulating in the blood (often called circulating tumor DNA, ctDNA). But HPV-DeepSeek goes further: it uses whole-genome sequencing of the entire HPV genome, along with advanced machine-learning approaches, to find even minute viral DNA fragments and other biomarkers.
In their studies, researchers tested archived blood samples collected years before diagnosis. Among 28 individuals who later developed HPV-associated head and neck cancer, HPV-DeepSeek detected HPV DNA in 22 of them, and all 28 healthy control samples tested negative. By improving the algorithm with machine learning, they achieved a detection rate of 27 out of 28 in follow-ups, including in samples as old as 10 years before diagnosis.
The test also shows very high specificity (i.e., few false positives) and sensitivity (i.e., true positives) reported as 99% for both inpatients at their first clinical presentation.
