Tumor intrinsic and extrinsic origins of circulating genomic and proteomic biomarkers for early detection of lung cancer

Fragmentomic-proteomic lung cancer test

Patient / Population Intervention / Exposure Comparison Outcome

In individuals with or without lung cancer, an integrated cfDNA fragmentomic-proteomic machine learning classifier detected lung cancer with an AUC of 0.91, validated in an independent prospective cohort.

N

718patients

Design

Translational diagnostic biomarker study of genome-wide cfDNA fragmentomes (low-coverage whole-genome sequencing of plasma) in 718 individuals with or without lung cancer, with serum proteins measured by multiplexed targeted immunoassays in a subset, including individuals from the prospective LUCAS diagnostic lung cancer study at Copenhagen University Hospital, and validation in 400 individuals from the prospective MICA study of patients suspected of having malignancy.

Endpoint

Detection of lung cancer by an integrated cfDNA fragmentomic-proteomic machine learning classifier, measured by area under the curve (AUC), alongside the tumour-intrinsic and tumour-extrinsic origins of the circulating signals.

ResultThe integrated cfDNA fragmentomic-proteomic machine learning classifier detected lung cancer with AUC=0.91 (95% CI 0.86-0.96) and was validated in the independent prospective MICA cohort. Tissue- and cell-type-specific analyses showed that both tumour-intrinsic changes and extrinsic host responses gave rise to cfDNA fragmentation profiles and protein biomarkers. Tumour-extrinsic signals were particularly informative in early-stage disease, where tumour-derived cfDNA and protein biomarkers had limited signal.

Velculescu V, Zavras J, Cristiano S, et al. Presented at ESMO Congress 2026, abstract LBA35, 25 October. Not yet published.

Discussion & critique

The study suggests that host-derived signals add to tumour-derived ones in the blood, and that combining cfDNA fragmentomics with proteomics gives good discrimination for lung cancer. The abstract reports only an AUC with no sensitivity, specificity, stage-specific performance or figures for the MICA validation cohort, and it does not say how large the protein subset was. This is diagnostic accuracy work with no clinical outcomes, and it does not show whether testing changes stage at diagnosis or mortality. The work was funded in part by a research grant from Delfi Diagnostics, and several authors hold interests in that company or in Artemyx.