Johns Hopkins researchers have validated an artificial intelligence-powered blood test that detected liver cancer in patients from two geographically and biologically distinct populations, while also identifying the biological signals that enable the test to recognize disease. The findings, published July 31 in Cell Press Blue, strengthen the case for genome-wide analysis of cell-free DNA as a noninvasive approach to cancer detection, particularly in populations where the causes of liver cancer differ substantially.
The test is based on DELFI, short for DNA Evaluation of Fragments for Early Interception. Rather than searching for a single mutation or protein, the platform examines millions of DNA fragments released into the bloodstream when cells die. These fragments, known as cell-free DNA, carry information about the tissues from which they originated and about the processes affecting those tissues. Cancer changes the way DNA is packaged and fragmented, producing genome-wide patterns, or “fragmentomes,” that can be detected computationally.
The new study evaluated a previously developed liver cancer classifier in blood samples from 377 people in Guatemala and Romania. Participants included individuals with and without hepatocellular carcinoma, the most common form of primary liver cancer. The two groups represented sharply different disease backgrounds. In Romania, liver cancer was frequently associated with viral hepatitis or alcohol-related liver disease. In Guatemala, many participants had metabolic liver disease, obesity and diabetes, and a substantial number had been exposed to aflatoxin, a naturally occurring toxin produced by certain fungi and strongly linked to liver cancer.
Despite these differences, the classifier consistently identified liver cancer in both populations. The researchers reported that combining fragmentome analysis with alpha-fetoprotein, or AFP, and basic clinical variables such as age and sex improved detection of both early- and late-stage disease compared with blood testing based on AFP alone. AFP is widely used in liver cancer surveillance, but it can remain normal in many patients with early tumors and can also be elevated in people who do not have cancer.
The findings are important because liver cancer is often diagnosed after curative treatment options have narrowed. Its global burden continues to grow as metabolic dysfunction-associated liver disease becomes more common, alongside persistent risks from viral hepatitis, alcohol exposure and environmental carcinogens. Ultrasound and AFP remain central to surveillance, but their performance can vary with patient characteristics, disease stage and the quality of imaging. A blood test capable of detecting biological changes before a tumor becomes readily visible could help expand screening, although prospective studies will be needed before the approach can be incorporated into routine care.
To understand why the test works, the investigators used a new analytical method called MethID. The method traces DNA fragments to their likely tissues of origin by examining methylation patterns, chemical marks that regulate gene activity and differ among cell types. This analysis showed that the DELFI signal is not generated solely by tumor DNA. Fragments from liver cells, blood-vessel cells and immune cells also contributed to the pattern, reflecting the surrounding tissue response to cancer.
That broader signal may help explain why fragmentome analysis can detect disease even when tumors shed relatively little DNA into the circulation. As liver cancer develops, malignant cells interact with the liver environment, alter blood-vessel behavior and recruit immune cells. Cell death and tissue remodeling then release DNA fragments with distinctive distribution, size and methylation characteristics. By integrating these signals across the genome, the algorithm can recognize a biological state associated with liver cancer rather than relying on one tumor-specific alteration.
The researchers also found molecular differences between the two populations. Participants from Guatemala showed a characteristic mutation pattern associated with aflatoxin exposure, providing a genomic indication of the environmental pathway that contributed to some cancers. Yet this regional signature did not prevent the overall fragmentome classifier from performing across both groups. The result suggests that the test captures fundamental features of liver cancer shared across populations while retaining the capacity to reveal causes and exposures that vary by region.
The work extends earlier research from the Johns Hopkins team showing that genome-wide cell-free DNA fragmentation patterns could identify liver fibrosis and cirrhosis, conditions that frequently precede liver cancer. Together, the studies point toward a potential continuum of blood-based monitoring, in which molecular signals could help identify chronic liver injury, advanced scarring and malignant transformation. The same technological framework is also being investigated for other diseases, including lung cancer, through the FirstLook Lung test developed with DELFI Diagnostics.
The researchers say the next steps include prospective clinical validation and refinement of multimodal testing that combines fragmentome signals, protein biomarkers and clinical risk factors. Such studies must establish how the test performs in routine surveillance, how often it produces false-positive results, and whether earlier detection ultimately improves survival. The investigators also disclosed financial and intellectual-property relationships involving DELFI Diagnostics and Artemyx, including company ownership, consulting roles and patents licensed from Johns Hopkins University. These relationships have been reviewed under the university’s conflict-of-interest policies.
Subject of Research: AI-powered liquid biopsy for early detection of hepatocellular carcinoma using cell-free DNA fragmentomics.
News Publication Date: July 31
Web References: Johns Hopkins Kimmel Cancer Center: https://www.hopkinsmedicine.org/kimmel-cancer-center; Johns Hopkins Bloomberg School of Public Health: https://publichealth.jhu.edu/; Johns Hopkins University School of Medicine: https://www.hopkinsmedicine.org/som; Cell Press: https://www.cell.com/cell-press-blue/home
Keywords: liver cancer, hepatocellular carcinoma, artificial intelligence, liquid biopsy, cell-free DNA, DELFI, fragmentomics, MethID, AFP, liver disease, cirrhosis, fibrosis, aflatoxin, viral hepatitis, early cancer detection
