Researchers at the Johns Hopkins Kimmel Cancer Center have developed a machine learning model, plasmaCHORD, that distinguishes tumor-derived DNA from non-cancerous biological noise in liquid biopsies. Published in Clinical Cancer Research, the study demonstrates that the tool improves the identification of clinically relevant mutations from 50% to 83% by analyzing DNA fragmentation patterns, helping clinicians avoid ineffective targeted therapies.
How does liquid biopsy technology filter out biological noise?
Liquid biopsies detect cell-free DNA (cfDNA) circulating in a patient’s blood to identify tumor mutations. However, these tests often capture mutations caused by clonal hematopoiesis, an aging-related process in white blood cells. According to Dr. Jenna Canzoniero, an Assistant Professor of Oncology at Johns Hopkins, these non-tumor mutations can mimic cancer-related genetic changes, potentially leading to the prescription of unnecessary or ineffective targeted drugs. The plasmaCHORD model addresses this by analyzing specific fragmentation profiles—the physical way DNA is “chopped up”—which differs between tumor cells and white blood cells.
Approximately one-third of mutations detected in liquid biopsies that do not have a prior tumor sample for comparison actually originate from white blood cells rather than the tumor itself, according to data from the Johns Hopkins Molecular Tumor Board.
What is the clinical impact of plasmaCHORD?
The model provides a scalable method to increase the precision of cancer treatment selection. In a validation cohort of 114 patients with breast, prostate, or non–small cell lung cancer, researchers found the model successfully identified the true source of mutations across different sequencing platforms. By filtering out white blood cell interference, the tool prevents clinicians from targeting “noise,” ensuring that mutation-targeted therapies are directed only at the patient’s actual tumor, as noted by senior study author Dr. Valsamo Anagnostou.
Comparison: Standard Sequencing vs. plasmaCHORD
| Method | Accuracy (Clinical Mutations) |
|---|---|
| Standard Liquid Biopsy | ~50% |
| plasmaCHORD Model | 83% |
What happens next for AI in oncology diagnostics?
The research team intends to refine plasmaCHORD to further improve performance metrics. While currently used as a proof-of-concept tool, the model is designed to be quickly scalable across various clinical environments. Future iterations may integrate more patient-specific variables, such as treatment history, to further increase the accuracy of distinguishing between tumor mutations and aging-related white blood cell variants. Clinicians at the Johns Hopkins Molecular Tumor Board are already using these insights to guide therapy decisions.
When reviewing liquid biopsy results, clinicians should consider the patient’s age and history of chemotherapy, as these factors increase the likelihood of clonal hematopoiesis mutations appearing in test reports.
Frequently Asked Questions
Why do liquid biopsies sometimes show false-positive results?
Liquid biopsies can detect mutations in white blood cells that are not related to cancer, especially in older patients or those previously treated with chemotherapy. These mutations are known as clonal hematopoiesis.
Is plasmaCHORD available for general clinical use?
The model is currently a research tool demonstrated to be clinically useful for the Johns Hopkins Molecular Tumor Board, with potential for broader clinical application as development continues.
Does the model require special blood tests?
No. According to the study, plasmaCHORD is designed to be applied to existing liquid biopsy sequencing data, making it a potentially cost-effective and scalable solution.
For more updates on advancements in cancer diagnostics, subscribe to our newsletter or explore our oncology research archives. Have questions about how AI is changing cancer care? Leave a comment below.
