The new tool, called meth-SemiCancer2, was developed by Joung Min Choi of Virginia Tech and Heejoon Chae of Sookmyung Women’s University in Seoul, and published in the journal BMC Bioinformatics. It builds on the team’s earlier meth-SemiCancer framework, which used semi-supervised learning to squeeze predictive power out of the vast quantities of unlabeled methylation data sitting in public repositories. The original model worked well, but its creators acknowledged a critical weakness: it did not explicitly correct for batch effects, the systematic technical differences that arise when samples are processed in different labs, on different instruments, or under different protocols.

Batch effects are the bane of large-scale genomics. In DNA methylation profiling, they can introduce artificial differences between groups of samples that dwarf the subtle biological signals researchers are trying to detect. When a machine learning model trained on one cohort is applied to another, these technical artifacts can masquerade as biological distinctions, producing spurious subtype boundaries that fail to replicate across studies. For a field where reproducibility is everything, this is more than an inconvenience — it can undermine the clinical utility of molecular classification altogether. Compounding the problem, methylation datasets are extraordinarily high-dimensional, with hundreds of thousands of CpG sites measured per sample, while genuinely subtype-labeled training data remain scarce and expensive to produce.

meth-SemiCancer2 tackles these intertwined challenges through a carefully orchestrated multi-phase pipeline that blends several of the most powerful ideas in modern machine learning. The process begins with pretraining on labeled source datasets, allowing the model to capture the subtype-specific methylation patterns that define known cancer classes. Next comes adversarial training, a domain adaptation technique in which the model learns to align the statistical distributions of the source data and an unlabeled target cohort. The goal is to make the model’s internal representations indistinguishable between the two domains, so that technical differences between cohorts are actively erased rather than passively tolerated.

The third phase introduces contrastive learning, a technique that has revolutionized representation learning across artificial intelligence. By teaching the model to pull together representations of samples that share underlying biology while pushing apart those that do not — regardless of which batch they originated from — contrastive learning strengthens the extraction of domain-invariant features. In other words, the model learns to focus on what makes a tumor biologically distinctive rather than on which laboratory processed the sample. Finally, the framework applies semi-supervised fine-tuning with subtype alignment, in which the model iteratively generates and refines pseudo-labels for the unlabeled target data, progressively sharpening the separation between subtypes across cohorts.

The researchers put this pipeline through a demanding battery of tests spanning seven cancer types: breast invasive carcinoma, colon adenocarcinoma, prostate adenocarcinoma, glioblastoma multiforme, renal cell carcinoma, thyroid carcinoma, and low-grade glioma. These cancers were chosen in part because they represent a spectrum of methylation-based classification challenges, from well-characterized subtypes to more diffuse molecular boundaries. Across this diverse panel, meth-SemiCancer2 consistently outperformed a formidable lineup of competitors, including the team’s own previous meth-SemiCancer model, a representative domain adaptation method, two state-of-the-art DNA methylation batch correction approaches, and conventional machine learning classifiers such as support vector machines, random forests, and logistic regression.

The evaluation metrics tell a story of both accuracy and robustness. On labeled source datasets, ten-fold cross-validation — a rigorous scheme in which the data are repeatedly partitioned so that every sample serves as a test case — confirmed the model’s superior accuracy and reliability in subtype prediction. Performance was assessed not only by raw classification accuracy but also by the Matthews correlation coefficient, a demanding measure that accounts for performance across all classes and resists inflation by class imbalance. When the trained models were applied to target cohorts from different sources, they generalized effectively, producing accurate subtype assignments with clearly delineated clusters, a result the authors visualized using dimensionality-reduction techniques that revealed clean, biologically coherent groupings rather than the fragmented, batch-driven patterns typical of uncorrected data.

Perhaps most striking for real-world applications is the framework’s resilience under data scarcity. The evaluations showed that meth-SemiCancer2 sustained stable performance even when only fractions of the source or target data were available. This matters enormously in clinical and research settings, where labeled methylation data are chronically limited and new cohorts often arrive with few or no subtype annotations. A model that can deliver dependable subtype calls from limited training material lowers the barrier for smaller institutions and rare-cancer studies, where assembling thousands of annotated samples is simply not feasible. It also suggests the framework could adapt more readily to emerging methylation reference datasets and to cancers for which subtype labels remain contested or incomplete.

The technical significance of the work lies in its integration of previously separate solutions. Batch correction tools have traditionally operated as preprocessing steps, adjusting the raw data before any classification takes place, while domain adaptation and semi-supervised learning have been treated as distinct modeling paradigms. By weaving adversarial domain alignment, contrastive representation learning, and pseudo-label-driven fine-tuning into a single coherent pipeline, meth-SemiCancer2 corrects technical noise and learns discriminative subtype representations simultaneously, rather than in isolation. This end-to-end design means the model can optimize its features specifically for the downstream task of subtyping, rather than for generic statistical properties of the data — a distinction that appears to explain its consistent edge over both dedicated batch correction methods and general-purpose classifiers.

The implications for precision oncology could be substantial. DNA methylation classifiers are already moving into clinical practice, most notably in brain tumor diagnostics, where methylation profiling has reshaped how gliomas and other central nervous system tumors are categorized. A framework that generalizes robustly across cohorts could help extend this success to other cancer types, harmonize classifications between institutions, and enable smaller centers to benefit from models trained on large reference cohorts elsewhere. The researchers have made meth-SemiCancer2 publicly accessible on GitHub, inviting the bioinformatics and oncology communities to adopt, test, and extend the tool. As methylation datasets continue to accumulate in repositories such as The Cancer Genome Atlas and the Gene Expression Omnibus, frameworks of this kind may prove essential for converting a flood of noisy, heterogeneous molecular data into the clean, reproducible subtype definitions on which the next generation of targeted cancer therapies will depend. The work was supported by the National Research Foundation of Korea through its Bio & Medical Technology Development Program, and the authors report no competing interests.

Subject of Research: A machine learning framework for cancer subtyping from DNA methylation profiles with batch effect correction

Article Title: meth-SemiCancer2: a cancer subtyping framework leveraging domain adaptation with contrastive learning for batch effect correction in DNA methylation profiles

Article References: Choi, J. M., & Chae, H. (2026). meth-SemiCancer2: a cancer subtyping framework leveraging domain adaptation with contrastive learning for batch effect correction in DNA methylation profiles. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06656-0

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06656-0

Keywords: DNA methylation, cancer subtyping, batch effect correction, contrastive learning, domain adaptation, semi-supervised learning, precision oncology, bioinformatics, epigenetics, deep learning, TCGA, machine learning

A team of researchers at the University of Naples Federico II—Antonio M. Rinaldi, Cristiano Russo and Cristian Tommasino—has now published a study in the journal Data Mining and Knowledge Discovery that tackles this problem not by making image-recognition models bigger, but by teaching them to see tissue the way biologists do: as networks of interacting cells and regions rather than as flat grids of pixels. Their framework, built on hierarchical graph neural networks, represents every detected cell nucleus as a node in a graph, every coherent tissue region as a node in a second, coarser graph, and then fuses the two levels into a single multi-scale structure that a neural network can reason over.

The motivation comes from a well-known weakness of the deep learning systems currently dominating computational pathology. Convolutional neural networks and Vision Transformers slice a digitized slide into fixed-size patches, extract local features from each patch, and aggregate the results into a final decision. This patch-based strategy delivers impressive accuracy, but it forces an awkward trade-off: high-resolution patches capture fine morphological detail but lose the surrounding context, while larger patches provide context at the cost of detail. More fundamentally, none of these architectures explicitly encodes the relationships between neighboring cells, the adjacency of tissue compartments, or the natural hierarchy that runs from nucleus to gland to tissue region—precisely the structures pathologists rely on when they reason through a diagnosis.

Graphs offer an escape from that limitation. In the Naples team’s pipeline, a Vision Transformer-based segmentation model first identifies individual nuclei within each tissue region of interest and produces pixel-wise embedding maps that encode local morphological and contextual information. Each nucleus becomes a node, described by features that can be handcrafted shape and texture descriptors, deep semantic vectors drawn from the segmentation network’s decoder, or a hybrid of both. Edges connect cells whose centroids lie close together, using a distance-constrained k-nearest-neighbor strategy in which the number of neighbors and a maximum distance threshold prevent biologically implausible long-range links. In parallel, a superpixel-based algorithm partitions the image into coherent tissue regions, and a Region Adjacency Graph links neighboring regions so that information can flow across shared biological interfaces.

The genuinely novel part is the hierarchical integration. A binary association matrix records which cell nodes belong to which tissue nodes, allowing structured information to propagate across scales. The researchers designed and compared three fusion architectures: a late-fusion model that processes cell and tissue graphs separately and concatenates the embeddings at the end; a layer-wise model that interleaves cellular and tissue information at every depth of the network; and a sequential model in which cellular information is fully processed first and then used to enrich tissue-level analysis, deliberately mimicking how a pathologist moves from the microscope to the wider field of view.

Which of these strategies wins was not obvious in advance, and the team treated the answer as an empirical question. They carried out a validation-driven Bayesian hyperparameter search across graph construction parameters, node feature configurations, pooling strategies, network depths, hidden dimensionality, and learning rates, always confining optimization to the validation split to avoid contaminating the test set. The systematic search covered k-nearest-neighbor values from 5 to 20, distance thresholds between 80 and 200 pixels, and graph network depths of two to four layers. Every reported result was averaged over multiple runs with different random seeds, with early stopping on validation loss to curb overfitting.

The evaluation was performed on BRACS, a clinically validated benchmark of breast carcinoma subtyping that stands out for its difficulty and realism. The dataset was collected at IRCCS Fondazione Pascale in Naples between 2019 and 2020, drawn from 189 patients aged 16 to 86, digitized at 40x magnification, and annotated by three board-certified pathologists using a collaborative review process to resolve disagreements. It contains 547 whole-slide images and 4,539 annotated regions spanning seven histological classes that trace the full progression of breast disease: normal tissue, pathological benign lesions, usual ductal hyperplasia, flat epithelial atypia, atypical ductal hyperplasia, ductal carcinoma in situ, and invasive carcinoma. Crucially, the data are split at the patient level, so no patient appears in more than one split—a design that prevents data leakage and reflects the reality of clinical deployment, where a model must generalize to people it has never seen.

The results, reported as weighted F1-scores to account for the dataset’s class imbalance, delivered a clear verdict: the sequential hierarchical fusion strategy proved most effective, achieving a weighted F1-score of 0.587 plus or minus 0.015 on the test set and outperforming both a reimplementation of the earlier HACT-NET baseline and the alternative fusion variants, while also showing lower variance. Along the way, the design-space analysis produced a set of practical lessons. Decoder-derived features from the segmentation network beat handcrafted descriptors; mean bounding-box pooling was more robust than contour-based pooling; the CellViT segmentation backbone yielded the most informative cellular features; and the best overall recipe paired Principal Neighborhood Aggregation layers for the cell graph with Graph Isomorphism Network layers for the tissue graph, which consistently outperformed PNA layers at the tissue level in both accuracy and stability.

Perhaps the most clinically significant finding is where the improvements concentrated. The hardest categories in breast histopathology are not the obvious ones—distinguishing invasive carcinoma from normal tissue is relatively straightforward—but the atypical and precancerous lesions that sit on the borderline between benign and malignant. Here the hierarchical framework shone, with per-class F1 gains of up to 0.187 for atypical ductal hyperplasia, the borderline lesion most associated with elevated carcinoma risk, along with consistent improvements for flat epithelial atypia and ductal carcinoma in situ. The confusion matrix showed that most remaining errors occurred between normal and pathological benign tissues, two non-malignant categories where a cautious misclassification carries far less clinical consequence than confusing an atypical lesion with a harmless one.

The authors are careful about the limits of their claims. The study is confined to supervised classification at the region-of-interest level on a single dataset; comparisons with large pathology foundation models and weakly supervised whole-slide pipelines are left to future work, as those address different problem formulations. The tissue regionalization relies on unsupervised superpixels, and segmentation and graph construction carry a nontrivial computational cost, although spatial indexing keeps neighborhood construction close to linear in practice for typical region sizes containing hundreds to a few thousand nuclei. The team also notes that while qualitative inspection of the learned embeddings showed atypical nuclei clustering with tissue regions exhibiting disrupted glandular organization—a pattern consistent with diagnostic criteria—expert-led validation of this interpretability remains follow-up work. Still, the framework’s modular design means it can be extended to whole-slide graphs, to other cancer types, and to emerging relational domains such as histogenomics and spatial transcriptomics. In an era when artificial intelligence is being asked to shoulder an ever-greater share of diagnostic labor, the Naples study makes a compelling argument that the path to trustworthy machine diagnosis runs not through bigger pixel-grinders, but through models that respect the biology written into the tissue itself.

Subject of Research: Hierarchical graph neural networks for classifying breast cancer subtypes from histopathological images

Article Title: Hierarchical graph networks for breast cancer subtype classification

Article References: Rinaldi, A. M., Russo, C., & Tommasino, C. (2026). Hierarchical graph networks for breast cancer subtype classification. Data Mining and Knowledge Discovery, 40(6), Article 104. https://doi.org/10.1007/s10618-026-01259-y

Image Credits: AI Generated

DOI: 10.1007/s10618-026-01259-y

Keywords: graph neural networks, breast cancer, digital pathology, histopathology, BRACS dataset, cell graphs, tissue graphs, deep learning, computer-aided diagnosis, hierarchical fusion, cancer subtyping, Vision Transformers