August 12, 2026
One tumor sample did not behave as expected.
The sample arrived at the biobank of Pomeranian Medical University in Szczecin, classified as glioblastoma, an aggressive and highly heterogeneous brain cancer. But when the team analyzed its DNA methylation profile with a machine learning model developed using Illumina methylation arrays, the model identified the sample as a pituitary adenoma—a mostly benign and biologically distinct tumor.
The scientists retraced the sample’s journey through the biobank. Two patients, one with glioblastoma and the other with an adenoma, had undergone procedures on the same day. The samples had been mislabeled.
“It shows that when we use these artificial intelligence–based solutions, we can use them not only to support the diagnostic process but for data sanity checks also,” says Jan Bińkowski, a bioinformatician and PhD candidate in the Independent Clinical Epigenetics Laboratory. “That’s absolutely amazing from the perspective of data quality assurance.”
The discovery highlights the power of DNA methylation patterns as biological fingerprints. Because different tumor types carry distinct epigenomic signatures, the team’s methylation-based classifier was able to recognize that the sample did not match its recorded diagnosis, ultimately revealing a labeling error.
Leading the work is Professor Tomasz K. Wojdacz, head of the Independent Clinical Epigenetics Laboratory and research director at the university’s Regional Centre for Digital Medicine. Wojdacz has spent approximately two decades investigating DNA methylation biomarkers and their potential applications in medicine. After conducting much of his career in Denmark and other European research institutions, he returned to Poland and established the laboratory in 2019.
His aim was to advance clinical epigenetics in a country where, at the time, most epigenetic research remained focused on basic biology.
“From the beginning, my first goal was to establish a team to work on the translational part of epigenetic research and then begin to consolidate scientists from Poland and Europe around the idea of clinical epigenetics,” Wojdacz says.
Under his leadership, the laboratory has grown into a multidisciplinary group working across genomics, epigenomics, transcriptomics, bioinformatics, and machine learning. It collaborates with two major regional hospitals, oncology departments across Poland, and other universities, while building a carefully managed biobank containing hundreds of tumor specimens.
Learning more from methylation
Although nearly every cell contains essentially the same DNA sequence, cells can use that information differently. Epigenetic modifications such as DNA methylation help regulate whether genes are active or silent. Cancer can disrupt these patterns, leaving molecular signatures that may reveal whether a tumor is present, what kind it is, where it originated, or how it might respond to treatment.
Wojdacz’s team initially used Illumina Infinium HumanMethylation450 BeadChip and Infinium MethylationEPIC arrays to build machine learning models for a research tool that classifies tumors. The resulting pan-cancer classifier was developed from thousands of samples and designed to detect cancer, distinguish tumor types, and identify a likely site of origin in metastatic disease. Bińkowski and Wojdacz reported the work in Genome Medicine.
In a prospective study, the classifier demonstrated the value of methylation arrays, which provide highly reproducible measurements at thousands of strategically selected sites across the genome. In the team’s glioblastoma study, array-based analysis confirmed the diagnosis in 15 of 16 samples and assigned precise molecular subtypes of glioblastoma to 14. These results highlight the potential of methylation-based analysis to deliver more accurate tumor characterization, enabling researchers to better understand disease biology and advance translational cancer research.
Arrays remain an efficient option for studying methylation across large sample cohorts. But Wojdacz saw the opportunity to go further.
“Look what we have done with only about 800,000 CpG sites from microarrays,” he says. “We have tumor and tissue classifiers based on this subset of CpG sites; we made major steps in deciphering cancer biology and have even been able to detect methylation changes in tumor free circulating DNA from liquid biopsies using this technology. It was a revolution in cancer research.”
The question was what might become possible when researchers could examine both the genome and the entire methylome—the genome-wide pattern of DNA methylation—in a single experiment.
Putting 5-base sequencing to the test
To explore that question, the group became an early user of Illumina 5-base sequencing. The technology generates information about genomic variation and DNA methylation together, allowing researchers to investigate the relationship between genetic changes and gene regulation without conducting separate assays.
The team evaluated 5-base sequencing using 16 fresh-frozen glioblastoma samples and compared the results with MethylationEPIC v2 arrays and Oxford Nanopore sequencing. Glioblastoma provided a demanding test: the tumors are heterogeneous, often contain necrotic fields, and can yield fragmented DNA.
Despite those challenges, 5-base sequencing produced stable yield with average genomic coverage of 56×. In the researchers’ head-to-head evaluation, it appeared to deliver the most accurate and uniform combined genomic and epigenomic profiling, while the methylation measurements showed strong agreement with EPIC v2 array results.
“The most important thing here is reproducibility, scalability, and data accuracy,” Bińkowski says. “For now, from my perspective, 5-base appears to be the most effective technology for combined genome and epigenome profiling.”
The ability to generate both kinds of information could help researchers ask questions that were previously examined separately. A single dataset may contain information about small variants, structural changes, copy-number alterations, methylation patterns, and the interaction between those signals.
For Wojdacz, that convergence represents a significant shift.
“Now we are beginning with a technology that, in one experiment, in a very cost-effective way, gives you the genome and methylome,” he says. “It’s just a matter of time before entire epigenomes and methylomes will be available—not only for cancer, but for other tissues as well.”
Building reproducibility into the workflow
Generating more data is only part of the challenge. To translate molecular biomarkers from research studies into dependable tools, researchers must also be able to reproduce results across laboratories, computing environments, and populations.
The Szczecin team uses Illumina BioInsight Platform Core (formerly Illumina Connected Analytics) as part of a framework for data processing and analysis. Standardized cloud workflows can reduce variations introduced when research groups use different software versions, pipelines, computing infrastructure, or analysis parameters.
“We can generate data in dozens of different laboratories and get the same results due to workflows standardization,” Bińkowski says. “From the perspective of research reproducibility, it’s absolutely pivotal.”
That consistency is especially important for machine learning. Models must be trained, tested, and validated on data that have been processed comparably; otherwise, technical differences can be mistaken for meaningful biological signals.
The group’s work remains focused on research, and substantial validation and regulatory review would be required before new classifiers or sequencing approaches could guide care. Still, the researchers envision a future in which comprehensive molecular profiling could help assess disease risk, detect cancer, identify its origin and subtype, predict treatment response, and monitor for residual disease or recurrence.
For Wojdacz, the excitement lies not only in collecting more information, but in generating it through repeatable processes that could eventually support precision medicine.
“We are on the brink of a revolution,” he says. “We have already witnessed the transformation of cancer research through standardized methylation array data. Now, 5-base sequencing delivers standardized methylation and genomic sequencing data together for the first time. That gives us a powerful new lens to understand disease biology and accelerate precision medicine.”
