Abstract
BACKGROUND: Lung cancer (LC) screening via low-dose computed tomography (LDCT) faces challenges including high false-positive rates and low patient compliance. Circulating tumor DNA (ctDNA)-based tests offer a minimally invasive alternative but are limited by high costs and low sensitivity, particularly in early-stage detection. This study introduces a cost-effective, shallow genome-wide sequencing approach for LC detection by profiling multiple cell-free DNA (cfDNA) signatures. METHODS: We developed a multimodal cfDNA assay with shallow sequencing coverage (0.5×) that integrates fragmentomic, nucleosome, end-motif, and copy number alteration analyses. A machine-learning model trained on a discovery cohort (99 LC patients, 168 healthy controls) and validated on an independent cohort (58 LC patients, 71 controls) demonstrated robust performance. RESULTS: The ensemble model exhibited outstanding performance, achieving an AUC of 0.97 and a specificity of 92% in both the discovery and validation cohorts, with sensitivities of 94% and 90%, respectively. Notably, it outperformed hotspot mutation-based assays and the multi-cancer SPOT-MAS assay in sensitivity across all LC stages. CONCLUSIONS: This assay provides a cost-effective, accurate, and minimally invasive method for LC detection, addressing the limitations of current screening methods. It represents a promising complementary tool to improve early detection and patient outcomes in LC.