The detection of circulating tumor DNA, which allows non-invasive tumor molecular profiling and disease follow-up, promises optimal and individualized management of patients with cancer. However, detecting small fractions of tumor DNA released when the tumor burden is reduced remains a challenge. We implemented a new highly sensitive strategy to detect base-pair resolution methylation patterns from plasma DNA and assessed the potential of hypomethylation of LINE-1 retrotransposons as a non-invasive multi-cancer detection biomarker. We have developed computational tools to accurately align sequencing data without a reference genome and applied prediction models, trained by machine learning algorithms, integrating patterns of methylation, overall and at the single molecule level. This assay, named DIAMOND (for Detection of Long Interspersed Nuclear Element Altered Methylation ON plasma DNA), showed powerful correct classification rates discriminating healthy and tumor plasmas from 6 types of cancers, including 3 at localized stages, in two independent cohorts (AUC = 88% to 100%, N = 747). To push the DIAMOND assay towards a clinically applicable test, we also demonstrated that DIAMOND data can be used to perform copy number alterations analysis which improves cancer detection. We integrated this analysis in a classifier providing ‘healthy' or ‘cancer' labels for each sample and reached a detection of 91% of true positives for all cancers together on the independent validation cohort.

This approach offers an optimized balance between the number of targeted regions and sequencing depth, which could extensively improve the sensitivity of ctDNA detection in a cost-effective manner and improve management of patients with cancer. This should lead to the development of more efficient non-invasive diagnostic tests adapted to all cancer patients, based on the universality of these factors.