Researchers at Johns Hopkins Kimmel Cancer Center have developed a groundbreaking machine learning model named plasmaCHORD, aimed at enhancing the accuracy of mutation identification in liquid biopsies for cancer diagnostics. This innovative tool significantly improves the clinical decision-making process by effectively distinguishing between cancer-derived mutations and those from other biological processes, which is crucial for targeted therapeutic interventions.
Liquid biopsies have revolutionized cancer diagnostics by allowing for the analysis of cell-free DNA (cfDNA) shed by tumors into the bloodstream. However, a major challenge has been the background noise from mutations in white blood cells due to clonal hematopoiesis, which complicates the identification of true tumor mutations. PlasmaCHORD addresses this issue by analyzing unique cfDNA fragmentation patterns, enabling it to differentiate between tumor-derived and hematopoietic mutations based on patient-specific factors.
The model was trained using data from 225 patients with various solid tumors and validated against matched tumor biopsies, showing a significant increase in accuracy from around 50% to 83% for clinically relevant mutations. Its robustness was further demonstrated in an independent cohort, maintaining high accuracy across different liquid biopsy technologies. This advancement not only enhances diagnostic precision but also optimizes treatment decisions, reducing the risk of unnecessary therapies.
As plasmaCHORD continues to evolve, plans for integrating additional genomic features and validating its effectiveness across diverse populations are underway. This collaborative effort reflects a strong commitment to advancing cancer care through artificial intelligence, positioning plasmaCHORD as a pivotal tool in the future of precision medicine.