Researchers at the Johns Hopkins Kimmel Cancer Center have developed and validated plasmaCHORD, a machine learning model designed to enhance the diagnostic accuracy of liquid biopsies. The study, published May 1 in Clinical Cancer Research, addresses a critical limitation in cancer genomics: the inability to distinguish tumor-derived mutations from biological noise originating in aging white blood cells. Liquid biopsies analyze cell-free DNA fragments circulating in the bloodstream to identify actionable mutations in solid tumors. However, these samples frequently contain mutations acquired through clonal hematopoiesis, an age-related process in white blood cells that can obscure true tumor signals. Misinterpreting these benign mutations as cancerous can lead clinicians to prescribe ineffective targeted therapies. To resolve this, the research team engineered plasmaCHORD to evaluate DNA fragmentation patterns, patient age, and specific gene characteristics. Tumor and white blood cell DNA are processed differently in the body, creating distinct fragmentation profiles. The model leverages these patterns to probabilistically assign each detected mutation to its correct biological source. The algorithm was trained on liquid biopsy data from 225 patients with breast, colorectal, esophageal, ovarian, or non-small cell lung cancer. Initial validation utilized matched sequencing of tumor and white blood cell samples to confirm mutation origins. Subsequent external testing across 114 patients with breast, prostate, or lung cancers, utilizing a different sequencing platform, demonstrated consistent performance. For clinically relevant mutations, plasmaCHORD increased correct classification accuracy from approximately 50 percent to 83 percent. Clinical utility was further established through the Johns Hopkins Molecular Tumor Board, where plasmaCHORD predictions successfully identified patients who would have otherwise received mismatched therapies. The findings indicate that integrating artificial intelligence into standard liquid biopsy workflows can significantly refine genomic profiling without requiring additional laboratory procedures. Lead investigator Jenna Canzoniero noted that the model provides a reliable tool for determining mutation origin when standard sequencing results are ambiguous. Senior author Valsamo Anagnostou emphasized that approximately one-third of mutations detected in tumor-naive liquid biopsies stem from white blood cells, making origin classification essential for precision oncology. Both researchers highlighted the platform potential for broad clinical deployment and ongoing model refinement. Collaborative contributions came from Vanderbilt University, LabCorp, the Netherlands Cancer Institute, and the University Medical Center Utrecht. The research team included Daniel Rabizadeh, Ilias Ziakas, Jaime Wehr, Archana Balan, Amna Jamali, Blair Landon, Lavanya Sivapalan, Susan Scot, Gavin Pereira, Vincent Lam, Christine Hann, Jessica Tao, Patrick Forde, Joseph Murray, Victor Velculescu, Jillian Phallen, and Robert Scharpf. The development marks a measurable advancement in computational pathology, positioning AI-driven liquid biopsy analysis as a scalable standard for patient stratification and treatment selection.