A groundbreaking advance in the field of oncology diagnostics heralds a new era for precision medicine. Researchers at the Johns Hopkins Kimmel Cancer Center have developed a sophisticated machine learning technique designed to dramatically improve the accuracy of mutation identification in liquid biopsy samples. This innovative tool, called plasmaCHORD, promises to significantly refine the clinical decision-making process by distinguishing cancer-derived mutations from those arising due to other biological processes, thereby allowing for more targeted and effective therapeutic interventions.
Liquid biopsies have emerged as a minimally invasive method to analyze cell-free DNA (cfDNA) fragments shed by tumors into the bloodstream. This approach has revolutionized cancer diagnostics by enabling continuous monitoring of tumor genomics without the need for traditional tissue biopsies. However, one grave challenge persists: the high background noise stemming from mutations accumulated in white blood cells through clonal hematopoiesis. This confounding signal often leads to ambiguity in discerning whether detected mutations truly originate from tumor cells or from aging-related alterations in blood cells, complicating therapeutic choices.
The plasmaCHORD model ingeniously tackles this challenge by scrutinizing distinct fragmentation patterns of cfDNA. Tumor-derived DNA fragments and those from white blood cells undergo differential cleavage processes, resulting in unique cfDNA fragmentation profiles. By leveraging these patterns alongside patient-specific variables such as age, gene involved, and mutation characteristics, the algorithm accurately predicts the source of each mutation. This nuanced analysis goes well beyond traditional sequencing, which often treats detected mutations without contextual origin differentiation.
Training plasmaCHORD involved the comprehensive analysis of liquid biopsy data from 225 patients afflicted with a variety of solid tumors including breast, colorectal, esophageal, ovarian, and non-small cell lung cancers. The model’s predictive power was rigorously validated using matched tumor biopsy and white blood cell sequencing, guaranteeing that the classification of mutation origins was grounded in unequivocal biological evidence. The initial results revealed a marked improvement in correctly identifying tumor mutations, setting a new benchmark for clinical molecular diagnostics.
To test the model’s robustness, the research team applied plasmaCHORD to an independent cohort comprising 114 patients with breast, prostate, or non-small cell lung cancers sourced from a different institution employing a different liquid biopsy sequencing technology. Remarkably, the tool maintained similar accuracy, distinguishing tumor mutations from hematopoietic mutations with high fidelity. PlasmaCHORD boosted the accuracy rates from a near-coin-flip 50% baseline to an impressive 83% for key mutations with clinical significance, underscoring its potential for widespread clinical adoption.
Clinically, this innovation transcends theoretical modeling, as demonstrated through its deployment within the Johns Hopkins Molecular Tumor Board. Integrating plasmaCHORD’s predictions enabled clinicians to circumvent the pitfall of selecting ineffective treatments driven by misattributed mutations. By ensuring that only tumor-specific mutations guide therapeutic decisions, the model optimizes patient outcomes, potentially reducing unnecessary drug exposure and associated toxicities. This synergy of artificial intelligence and clinical oncology epitomizes the future of personalized cancer treatment.
One-third of mutations identified in tumor-naive liquid biopsies are believed to stem from white blood cells — a statistic that has long hindered the clinician’s ability to tailor precision therapies based on liquid biopsy results alone. By incorporating plasmaCHORD into the diagnostic workflow, oncologists gain an unprecedented clarity to confidently target mutations that genuinely underpin the patient’s malignancy, thereby reinforcing the integral link between molecular profiling and precision therapy.
The impetus behind plasmaCHORD is grounded in a deep understanding of cfDNA biology. DNA fragments circulating in the blood originate from multiple physiological processes, each imparting distinct fragmentation signatures. Tumor cells often release DNA with specific sizes and cleavage patterns due to apoptosis and necrosis mechanisms distinct from those active in hematopoietic cells. Capturing these fragmentation nuances enables plasmaCHORD to function as a molecular detective, distinguishing subtle signals in a complex cfDNA milieu.
The Johns Hopkins research team led by co-authors Jenna Canzoniero, M.D., M.S., and Valsamo Anagnostou, M.D., Ph.D., envisions that future iterations of plasmaCHORD will refine predictive accuracy even further. Plans are underway to integrate additional genomic and epigenomic features, as well as to validate the model across larger and more diverse populations. Such advances will pave the way for plasmaCHORD to be seamlessly embedded into routine clinical workflows and multi-institutional cancer genomic databases.
Collaborative efforts across multiple academic and industry institutions, including Vanderbilt University, LabCorp, and the Netherlands Cancer Institute, underscore the broad interest and trust in plasmaCHORD’s transformative potential. Funding from prestigious bodies such as the National Cancer Institute and the Department of Defense reflects the critical importance of this research to national cancer control priorities and the future of cancer care innovation.
The advent of plasmaCHORD exemplifies how artificial intelligence can unravel complex biological signals obscured by noise, delivering enhanced diagnostic precision. As liquid biopsies continue to gain prominence, tools like plasmaCHORD will be instrumental not only in honing treatment selection for individual patients but also in accelerating research toward overcoming cancer’s evolving genetic landscape.
In summary, plasmaCHORD stands as a beacon of progress in the quest to decode the biological origin of cfDNA mutations. By marrying novel machine learning algorithms with deep molecular understanding, it elevates liquid biopsy from a promising concept to a powerful clinical utility, allowing oncologists to focus precisely on the genetic hallmarks of tumors and tailor therapies with newfound confidence and accuracy.
Subject of Research: Application of machine learning to improve mutation source identification in liquid biopsies for cancer diagnosis and treatment.
Article Title: Development of an artificial intelligence method to accurately characterize mutations in liquid biopsies
News Publication Date: May 1, 2024
Web References:
https://doi.org/10.1158/1078-0432.CCR-25-0976
https://www.hopkinsmedicine.org/kimmel-cancer-center
Image Credits: Valsamo Anagnostou/ChatGPT
Keywords: Liquid biopsy, cell-free DNA, plasmaCHORD, machine learning, clonal hematopoiesis, cancer diagnostics, mutation characterization, precision oncology, molecular tumor board
