TORONTO, July 5th 2022 – Lung cancer is a leading cause of cancer-related mortality worldwide, making up 25% of all cancer deaths. Early diagnosis benefits lung cancer patients with better survival. Radiological approaches such as the low-dose computed tomography (LDCT) scan have been recommended for early screening purposes but show limited application. The liquid biopsy-based cell-free DNA (cfDNA) analysis has emerged recently as a promising non-invasive approach to the clinical practice of disease detection. A study from our research program DECIPHER (Detecting Early Cancer by Inspecting ctDNA Features) published in the journal of Ebiomedicine, led by the Chinese Academy of Medical Sciences, Peking Union Medical College, and Geneseeq Technology Inc., has developed a new machine learning model for sensitive detection of stage I lung adenocarcinoma (LUAD) using cfDNA breakpoint motif feature.
This study enrolled 292 stage I LUAD patients from three medical centers in China as well as 230 healthy volunteers. The LUAD patients included invasive adenocarcinoma (ADC) and minimally invasive adenocarcinoma (MIA). The cfDNA was prepared from the plasma samples, followed by whole-genome sequencing (WGS). Multiple cfDNA fragmentomic motif features and machine learning algorithms were investigated to select the best model. During this process, the participants from Center I were randomly assigned to the training and internal validation cohorts for model construction and cutoff determination. The cancer patients from Centers II and III were assigned to the external validation cohort with 40 healthy controls for independent validation.
A novel 6bp-breakpoint-motif feature using the logistic regression model reached 98.0% sensitivity and 94.7% specificity in the internal validation cohort (Area Under the Curve AUC: 0.985), and 92.5% sensitivity and 90.0% specificity in the external validation cohort (AUC: 0.954), consistently outperforming other cfDNA-based methods for stage I lung adenocarcinoma detection. Notably, this assay is sensitive for early-stage (e.g., 100% sensitivity for MIA) and <1 cm (92.9%-97.7% sensitivity) tumors. The predictive power remained high with 0.5× WGS (AUC: 0.977 and 0.931 for internal and external cohorts). These results justified the cfDNA breakpoint motif-based machine learning model for detecting early-stage LUAD, especially the MIA and very small-size tumors.
“Noninvasive detection of early-stage lung cancer using plasma cfDNA has attracted increasing attention and is still in progress for leveraging its performance. Optimal cfDNA features and machine learning algorithms could improve the prediction power and need to be intensively tested in the clinical settings”, says Dr. Hua Bao, author and the Associate Dean of Geneseeq Research Institute.