Statistical Method Improves Detection of Low-Level Cancer DNA in Blood Samples

Tumor DNA levels often fall during effective therapy, making serial monitoring more difficult precisely when clinicians need clearer signals. Cost-efficient sequencing strategies can further introduce noise that obscures clinically relevant information. New findings demonstrate an analysis method that recovers tumor features from blood samples containing as little as 5% cancer DNA.

Researchers at Chalmers University of Technology and the University of Gothenburg have developed BayesCNA, a statistical approach designed for liquid biopsy analysis when cancer DNA is scarce. The method can analyze samples containing about 5% cancer DNA, compared with the 15–20% typically required by current techniques. By enabling evaluation of lower-quality samples, BayesCNA is intended to improve insights into tumor composition over time.

BayesCNA operates on data from low-pass whole-genome sequencing (low-pass WGS), a cost-advantaged technique that offers a broad view of DNA structure but with limited resolution due to sparse coverage. The new algorithm amplifies weak tumor-specific signals embedded in predominantly healthy DNA, extracting information that would otherwise be missed. The team initially evaluated machine learning approaches but found that classical statistical modeling delivered superior performance for this task.

Details of the work appear in Briefings in Bioinformatics as an experimental study published on March 16, 2026, under the title “Sensitive detection of copy number alterations in low-pass liquid biopsy sequencing data.” According to the institutions, the method can provide more detailed readouts of tumor composition from routine blood draws, supporting closer assessment of how cancer evolves during treatment. The researchers note that such analysis could complement intervals between invasive procedures by revealing changes that occur between treatment sessions.

The group outlines plans to further interrogate the tumor features recoverable by BayesCNA and to develop companion methods to identify characteristics linked to treatment response. They also express interest in broader collaborations to advance adoption within the research community.

“When the treatment is effective, the amount of cancer DNA in the blood drops significantly. This makes it more difficult both to detect the cancer and to monitor how it changes. It is important to be able to analyze samples containing low levels of cancer DNA to gain a clearer picture of how a patient responds to treatment,” said Eszter Lakatos, Assistant Professor in the Department of Mathematical Sciences at Chalmers and the University of Gothenburg.

“Nowadays, machine learning is used to solve a great many problems, and we tried those methods first. But, to our surprise, it turned out that classical statistics worked better in this case, which was particularly pleasing to us mathematicians and statisticians,” said Lotta Eriksson, doctoral student in the Department of Mathematical Sciences at Chalmers University of Technology and the University of Gothenburg.

Related Links
Chalmers University of Technology
University of Gothenburg