New blood test spots four cancers and other diseases by stripping away healthy DNA noise

By Hugo Francisco de SouzaReviewed by Susha Cheriyedath, M.Sc.Apr 7 2026

By cutting through the overwhelming background of healthy blood DNA, MethylScan gives clinicians a clearer view of cancer, liver disease, and tissue damage from a single blood sample.

Study: Toward the simultaneous detection of multiple diseases with a highly cost-effective cell-free DNA methylome test. Image Credit: 3dMediSphere / Shutterstock

In a recent study published in the journal PNAS, researchers introduced “MethylScan”, a novel method for cost-effective profiling of the cell-free DNA (cfDNA) methylome. The publication details how specialized enzymes can be used to facilitate the depletion of healthy background DNA, thereby enabling MethylScan to specifically enrich rare disease signals and substantially reduce sequencing costs.

The method was validated in a cohort of 1,061 individuals, with study assays demonstrating MethylScan’s high accuracy in detecting four major cancer types, classifying liver diseases, and identifying organ damage, marking progress toward holistic, pan-disease diagnostics. By significantly reducing sequencing costs while maintaining high analytical sensitivity, MethylScan establishes a viable framework for comprehensive, noninvasive health monitoring.

Background

Plasma cell-free DNA (cfDNA) is a heterogeneous mixture of fragments released from various tissues during cell death and has been previously used to provide clinicians with a noninvasive window into organ health. The technique’s advantages lie in its mechanistic underpinning. Unlike the genome, which remains largely static, the DNA methylome is tissue-specific and undergoes dynamic alterations in response to disease etiology, providing information on both tissue origin and disease state.

Unfortunately, the vast majority of circulating cfDNA (estimated at ~85%) originates from healthy hematopoietic cells. Studies have shown that during the early stages of oncogenesis, tumor-derived cfDNA may constitute less than 0.1% of the total sample. Consequently, to detect these trace signals, researchers typically require sequencing coverage exceeding 1,000x at specific target loci, making genome-wide approaches financially impractical for population-level screening.

About the study

The present study aimed to address this challenge by developing a novel methodological approach (named “MethylScan”) that selectively removes the dominant blood-derived background, effectively increasing the signal-to-noise ratio for disease-specific hypermethylation markers.

The MethylScan method uses a specialized experimental protocol and a custom-designed capture panel. The method’s core innovation lies in the use of Methylation-Sensitive Restriction Enzymes (MSREs), specifically: 1. HpaII (recognition site CCGG) and 2. HhaI (recognition site GCGC). These enzymes cleave DNA only when the recognition sites are hypomethylated, a state characteristic of the background cfDNA from white blood cells in targeted "panel regions".

The custom-designed capture panel comprised 154,028 target regions (each 120-bp in length), totaling 1,600,725 CpG sites. These regions were selected based on consistent hypomethylation across 48 white blood cell and 30 healthy plasma samples.

To normalize read counts (and thereby facilitate quantification), the study further included 300 control regions devoid of CpG or MSRE sites. Finally, the clinical validation of MethylScan’s accuracy and specificity was conducted by testing plasma (n = 1,061) and tissue (n = 899) samples collected for separate discovery and validation analyses.

This sample cohort included 460 cancer patients across four cancer types (liver, lung, ovarian, and stomach) alongside 601 noncancer individuals, including general hospital visitors and people at high risk because of chronic liver disease or benign lung nodules.

To optimize model performance (especially to prevent overfitting), the study applied singular value decomposition (SVD) for dimensionality reduction before training Linear Support Vector Machine (LSVM) classifiers.

Study findings

Analytical evaluation via a liver tumor DNA dilution series demonstrated that MethylScan can detect cancer signals at tumor fractions as low as 0.05% (Student’s t-test p = 0.000058) with a strong linear correlation between expected and estimated fractions (R2 = 0.983).

Notably, compared to samples without MSRE digestion, the MethylScan protocol achieved 5.1 times higher coverage in control regions per million paired-end reads, highlighting its cost-effectiveness.

In clinical applications, the assay model achieved an average area under the receiver operating characteristic curve (AUROC) of 0.938 (95% CI: 0.920–0.954) for multicancer detection across these four cancers. At a specificity of 98.0%, it achieved 63.3% sensitivity across all cancer stages and 55.3% in early-stage (Stage I/II) cancers. For Stage I specifically, the AUROC was 0.906 with 54.4% sensitivity.

MethylScan’s accuracy in distinguishing the four cancer types was 91.7% (95% CI: 87.9–94.6%) for all stages and 89.8% for early stages. The method also classified a subset of high-risk patients with Hepatitis B, Hepatitis C, alcoholic liver disease, and MASLD with an overall accuracy of 84.7% (95% CI: 76.0–91.2%).

Patients with liver cancer or liver disease showed significantly elevated liver-derived cfDNA fractions compared to general hospital visitors (p = 4.1 x 10-16 for cancer; p = 5.6 x 10-3 for disease), highlighting tissue deconvolution. Finally, using methylation patterns associated with ancestry, the assay predicted race with 97.1% accuracy (95% CI: 94.6–98.7%) in analyses limited to White and Asian participants because other groups were too small for reliable cross-validation.

Conclusions

MethylScan represents a significant advancement in epigenomic diagnostics by overcoming the prohibitive costs of deep sequencing through MSRE-mediated background depletion. Its ability to capture rich, genome-wide information from a single 10 ng cfDNA sample enables continuous data stacking, which the authors note could support future machine learning and AI applications.

The method, therefore, represents a versatile platform that could help move the field toward a holistic diagnostic paradigm in which multiple health conditions can be identified and monitored simultaneously and cost-effectively.