PubMed Health⌕ Search

SEARCH · PubMed Health

Results for “Contrastive learning”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 19 recordsLinked to original sources
PubMed2026

ORBIT: Oncogenic Representation Learning via Bi-Prototype Contrastive Learning in Hyperbolic Space for cancer driver gene identification.

Accurate identification of cancer driver genes is crucial for precision oncology but remains challenging due to the complexity of integrating heterogeneous data and modeling dynamic biological systems. To address these limitations, we propose ORBIT (Oncogenic Representation Learning via Bi-Prototype Contrastive Learning in Hyperbolic Space). Our framework synergistically fuses multi-omics profiles with functional network data using a context-adaptive graph reweighting mechanism to capture cancer-specific dynamics. The model employs a bi-prototype contrastive learning strategy within hyperbolic space, which aligns gene representations around distinct driver and non-driver semantic anchors while preserving the intrinsic hierarchy of biological networks. Comprehensive evaluations demonstrate that ORBIT achieves highly competitive stability in pan-cancer analysis while consistently outperforming state-of-the-art methods in cancer-specific predictions. Furthermore, functional enrichment analysis confirms that the model effectively segregates core cancer pathways, and drug sensitivity profiling validates the clinical relevance of the identified drivers. By integrating hyperbolic geometry with context-adaptive learning, ORBIT offers a robust and interpretable paradigm for precision medicine. The source codes and datasets are publicly accessible at https://github.com/spcho-dev/ORBIT.

Humans↗
PubMed2026

Dual-contrastive learning for spatial domain identification in spatial transcriptomics with STAMGC.

Spatial transcriptomics (STs) have become a valuable approach for understanding the growth and development of organisms. Despite the recent emergence of numerous ST models, accurately identifying spatial domains remains challenging owing to the trade-off between preserving local details and reducing noise. Here, we introduce STAMGC, which is a dual-contrastive learning framework built upon graph convolutional networks. This model leverages regional and topological contrastive learning to jointly optimize the model, effectively reducing the noise in spatial domain identification and enhancing the extraction of detailed features. In this study, Gaussian smoothing, originally developed in the image processing field, is introduced to process ST data, providing a foundation for region-level contrastive learning by mitigating spatial discontinuities of gene expression signals. Experimental results indicate that STAMGC outperforms existing methods across multiple data sets according to comprehensive evaluations. Furthermore, STAMGC not only identifies finer structures in the mouse brain but also brings new discoveries for human breast cancer research.

Journal Article↗
PubMed2025

CDACHIE: chromatin domain annotation by integrating chromatin interaction and epigenomic data with contrastive learning.

MOTIVATION: Chromatin domain annotation identifies functional genomic regions, such as active and inactive zones, based on epigenomic features like histone modifications, DNA methylation, and chromatin accessibility. While recent methods have utilized both chromatin interaction data (e.g. Hi-C) and epigenomic data, they often overlook the direct relationship between these data types. RESULTS: In this study, we introduce Chromatin Domain Annotation using Contrastive Learning for Hi-C and Epigenomic Data (CDACHIE), a method for identifying chromatin domains from Hi-C and epigenomic data. Our approach leverages contrastive learning to generate aligned representative vectors for both data types at each genomic bin. The concatenated vectors are then clustered using K-means to classify distinct chromatin domain types. CDACHIE achieves superior performance in Variance Explained, evaluated across gene expression, replication timing, and ChIA-PET data. This highlights its robust ability to integrate semantic associations between Hi-C and epigenomic features within the embedding space. AVAILABILITY AND IMPLEMENTATION: The source code is available at GitHub: https://github.com/maruyama-lab-design/CDACHIE. An archival snapshot of the code used in this study is available on Zenodo: https://doi.org/10.5281/zenodo.15751780.

Chromatin↗
PubMed2026

scFANCL: Dual contrastive learning with false-negative correction at cell level for single-cell RNA-seq clustering.

BACKGROUND: Single-cell RNA sequencing (scRNA-seq) enables cellular characterization at single-cell resolution. However, its high dimensionality, sparsity, and noise make clustering challenging. Approaches utilizing contrastive learning and data augmentation have been introduced to improve representation quality for scRNA-seq clustering. In particular, dual contrastive frameworks combining instance- and cluster-level objectives can capture both cell-cell similarities and inter-cluster variations. However, existing dual contrastive frameworks focus primarily on discrete cluster boundaries, neglecting the biological continuity inherent in scRNA-seq data. METHODS: We propose scFANCL, a dual contrastive framework designed to capture biological continuity in scRNA data. Rather than treating all non-augmented samples as negatives, scFANCL applies a cosine-similarity-based threshold to exclude cells of the same type from the negative pool, preserving continuous transcriptional relationships among them while maintaining inter-cluster separation. RESULTS: Extensive experiments across seven publicly available scRNA-seq datasets demonstrated that scFANCL achieves competitive clustering performance compared with existing baseline methods, consistently yielding high ARI and NMI scores across datasets of varying size and complexity. Ablation studies further confirmed the contribution of the false negative filtering component, showing measurable improvements over variants without filtering. Downstream analyses further suggest that the learned embeddings may reflect biologically meaningful transcriptional transitions, including continuous differentiation trajectories within related cell types. The source code is available at https://github.com/mjuailab/scFANCL . CONCLUSIONS: scFANCL addresses a key limitation of conventional contrastive learning by applying a cosine-similarity-based threshold to exclude cells of the same type from the negative pool, thereby preserving biological continuity within cell types while maintaining inter-cluster separation. Evaluations across seven benchmark scRNA-seq datasets demonstrate competitive clustering performance, with learned embeddings capturing biologically meaningful transcriptional structure and characteristics of rare cell populations.

Clustering Algorithms↗
PubMed2025

A Graph Contrastive Learning Method for Enhancing Genome Recovery in Complex Microbial Communities.

Accurate genome binning is essential for resolving microbial community structure and functional potential from metagenomic data. However, existing approaches-primarily reliant on tetranucleotide frequency (TNF) and abundance profiles-often perform sub-optimally in the face of complex community compositions, low-abundance taxa, and long-read sequencing datasets. To address these limitations, we present MBGCCA, a novel metagenomic binning framework that synergistically integrates graph neural networks (GNNs), contrastive learning, and information-theoretic regularization to enhance binning accuracy, robustness, and biological coherence. MBGCCA operates in two stages: (1) multimodal information integration, where TNF and abundance profiles are fused via a deep neural network trained using a multi-view contrastive loss, and (2) self-supervised graph representation learning, which leverages assembly graph topology to refine contig embeddings. The contrastive learning objective follows the InfoMax principle by maximizing mutual information across augmented views and modalities, encouraging the model to extract globally consistent and high-information representations. By aligning perturbed graph views while preserving topological structure, MBGCCA effectively captures both global genomic characteristics and local contig relationships. Comprehensive evaluations using both synthetic and real-world datasets-including wastewater and soil microbiomes-demonstrate that MBGCCA consistently outperforms state-of-the-art binning methods, particularly in challenging scenarios marked by sparse data and high community complexity. These results highlight the value of entropy-aware, topology-preserving learning for advancing metagenomic genome reconstruction.

canonical correlation analysis↗
PubMed2026

How Not to be Seen: Predicting Unseen Enzyme Functions using Contrastive Learning.

MOTIVATION: Predicting enzyme function from its sequence is still an unsolved problem in the life sciences. Moreover, with the explosion of annotated genome data, we are inundated with potential enzymatic sequences that have not yet been biochemically characterized. While it is not possible to assign a not-yet-existing label to such a sequence, there is high value in placing the sequence as accurately as possible in known function space. Doing so can help provide more accurate falsifiable hypotheses for experimentalists wishing to characterize enzymes from specific functional families. RESULTS: Here we present a contrastive learning algorithm for predicting enzyme function from sequence. Our method, EnzPlacer, predicts the third, second, and first EC numbers for a protein whose fourth EC number is not in the training corpus. This novel prediction mechanism accurately places a protein sequence within a narrowed-down functional context, even if the precise function remains unknown. AVAILABILITY: EnzPlacer is available from https://github.com/drxiangma/EnzPlacer under a GPL3 license.

Contrastive learning↗
PubMed2025

ASGCL: Adaptive Sparse Mapping-based graph contrastive learning network for cancer drug response prediction.

Personalized cancer drug treatment is emerging as a frontier issue in modern medical research. Considering the genomic differences among cancer patients, determining the most effective drug treatment plan is a complex and crucial task. In response to these challenges, this study introduces the Adaptive Sparse Graph Contrastive Learning Network (ASGCL), an innovative approach to unraveling latent interactions in the complex context of cancer cell lines and drugs. The core of ASGCL is the GraphMorpher module, an innovative component that enhances the input graph structure via strategic node attribute masking and topological pruning. By contrasting the augmented graph with the original input, the model delineates distinct positive and negative sample sets at both node and graph levels. This dual-level contrastive approach significantly amplifies the model's discriminatory prowess in identifying nuanced drug responses. Leveraging a synergistic combination of supervised and contrastive loss, ASGCL accomplishes end-to-end learning of feature representations, substantially outperforming existing methodologies. Comprehensive ablation studies underscore the efficacy of each component, corroborating the model's robustness. Experimental evaluations further illuminate ASGCL's proficiency in predicting drug responses, offering a potent tool for guiding clinical decision-making in cancer therapy.

Humans↗
PubMed2026

How not to be seen: predicting unseen enzyme functions using contrastive learning.

MOTIVATION: Predicting enzyme function from its sequence is still an unsolved problem in the life sciences. Moreover, with the explosion of annotated genome data, we are inundated with potential enzymatic sequences that have not yet been biochemically characterized. While it is not possible to assign a not-yet-existing label to such a sequence, there is high value in placing the sequence as accurately as possible in known function space. Doing so can help provide more accurate falsifiable hypotheses for experimentalists wishing to characterize enzymes from specific functional families. RESULTS: Here we present a contrastive learning algorithm for predicting enzyme function from sequence. Our method, EnzPlacer, predicts the third, second, and first EC numbers for a protein whose fourth EC number is not in the training corpus. This novel prediction mechanism accurately places a protein sequence within a narrowed-down functional context, even if the precise function remains unknown. AVAILABILITY AND IMPLEMENTATION: EnzPlacer and data is available at https://github.com/drxiangma/EnzPlacer under a GPL3 license.

Enzymes↗
PubMed1994

[Do radiologists develop perceptual learning contrast sensitivity?].

A slight difference in brightness between objects close to each other and with no clear-cut outlines separating them can be recognized by the visual function called contrast sensitivity. This function is particularly developed in the radiologist, whose task is to analyze images in many shades of grey and with no clear-cut outlines, due to kinetic and radiogeometrical shading. Assuming that professional habit might develop this function, the authors compared contrast sensitivity in a group of 26 radiologists with contrast sensitivity in a control group (30 non-radiologists). The Vistech VTCS 6500 test was chosen for the task because of its practicality and reliability. The test consists in the recognition of the orientation of 5 series of alternate bright and dark lines differencing in thickness and contrast. The results obtained in the two groups were studied and compared with the statistical test of the analysis of variance, the T-test by Student and the U-test by Mann-Whitney. Contrast sensitivity was surprisingly higher in the radiologists group, in 4 of 5 series of images with bright and dark lines, than in controls and the difference was statistically significant: p < 0.001 with the test and the U-test. A deviation standard increase in contrast sensitivity was found also in radiologists with short-term professional activity (a few years). To explain these results, two hypotheses one suggested by the authors: radiologists pay greater attention to image details, because of their profession; perceptual learning--that is an improvement in the image-research cortical function--might occur, because of constant training. In the radiologist to conclude repeated stimulation of research capabilities in low-contrast images is thought to improve the efficiency of this cortical function.

Adult↗
PubMed2026

mmContext: an open framework for multimodal contrastive learning of omics and text data.

SUMMARY: Multimodal approaches are increasingly leveraged for integrating omics data with textual biological knowledge. Yet there is still no accessible, standardized framework that enables systematic comparison of omics representations with different text encoders within a unified workflow. We present mmContext, a lightweight and extensible multimodal embedding framework built on top of the open-source Sentence Transformers library. The software allows researchers to train or apply models that jointly embed omics and text data using any numeric representation stored in an AnnData.obsm layer and any text encoder available in Hugging Face. mmContext supports integration of diverse biological text sources and provides pipelines for training, evaluation, and data preparation. We train and evaluate models for a RNA-Seq and text integration task, and demonstrate their utility through zero-shot classification of cell types and diseases across four independent datasets. By releasing all models, datasets, and tutorials openly, mmContext enables reproducible and accessible multimodal learning for omics-text integration. AVAILABILITY AND IMPLEMENTATION: Pretrained checkpoints and full source code for our custom MMContextEncoder are available on Hugging Face huggingface.co/jo-mengr. The Python package github.com/mengerj/mmcontext provides the model implementation and training and evaluation scripts for custom training. The releases for the publication can be accessed via zenodo: adata_hf_datasets: doi.org/10.5281/zenodo.19185217 and mmContext: doi.org/10.5281/zenodo.19185493.

Computational Biology↗
PubMed2026

Data distribution impacts the performance and generalisability of contrastive learning-based foundation models of electrocardiograms.

Contrastive learning is a widely adopted self-supervised pretraining strategy, yet its dependence on cohort composition remains underexplored. We present Contrasting by Augmented Patient Electrocardiograms (CAPE) foundation model and pretrain on four cohorts (n&#x2009;=&#x2009;5,203,269), from diverse populations across three continents (North America, South America, Asia). We systematically assess how cohort demographics, health status, and population diversity influence the downstream performance for prediction tasks also including two additional cohorts from another continent (Europe). We find that downstream performance depends on the distributional properties of the pretraining cohort, including demographics and health status. Moreover, while pretraining with a multi-centre, demographically diverse cohort improves in-distribution accuracy, it reduces out-of-distribution (OOD) generalisation of our contrastive approach by encoding cohort-specific artifacts. To address this, we propose the In-Distribution Batch (IDB) strategy, which preserves intra-cohort consistency during pretraining, discourages learning of spurious cohort-specific features, and instead promotes clinically meaningful variability within cohorts. This leads to improved out-of-distribution robustness, with gains of 9-40% in downstream label prediction performance. This work provides insights into pretraining strategies for more clinically deployable and generalisable foundation models.

Journal Article↗
PubMed2005

Learning to identify contrast-defined letters in peripheral vision.

Performance for identifying luminance-defined letters in peripheral vision improves with training. The purpose of the present study was to examine whether performance for identifying contrast-defined letters also improves with training in peripheral vision, and whether any improvement transfers to luminance-defined letters. Eight observers were trained to identify contrast-defined letters presented singly at 10 degrees eccentricity in the inferior visual field. Before and after training, we measured observers' thresholds for identifying luminance-defined and contrast-defined letters, embedded within a field of white luminance noise (maximum luminance contrast=0, 0.25, and 0.5), at the same eccentric location. Each training session consisted of 10 blocks (100 trials per block) of identifying contrast-defined letters at a background noise contrast of 0.5. Letters (x-height=4.2 degrees) were the 26 lowercase letters of the Times-Roman alphabet. Luminance-defined letters were generated by introducing a luminance difference between the stimulus letter and its mid-gray background. The background noise covered both the letter and its background. Contrast-defined letters were generated by introducing a differential noise contrast between the group of pixels that made up the stimulus letter and the group of pixels that made up the background. Following training, observers showed a significant reduction in threshold for identifying contrast-defined letters (p<0.0001). Averaged across observers and background noise contrasts, the reduction was 25.8%, with the greatest reduction (32%) occurring at the trained background noise contrast. There was virtually no transfer of improvement to luminance-defined letters, or to an untrained letter size (2 x original), or an untrained retinal location (10 degrees superior field). In contrast, learning transferred completely to the untrained contralateral eye. Our results show that training improves performance for identifying contrast-defined letters in peripheral vision. This perceptual learning effect seems to be stimulus-specific, as it shows no transfer to the identification of luminance-defined letters. The complete interocular transfer, and the retinotopic (retinal location) and size specificity of the learning effect are consistent with the properties of neurons in early visual area V2.

Adult↗
PubMed2003

Equivalence of backpropagation and contrastive Hebbian learning in a layered network.

Backpropagation and contrastive Hebbian learning are two methods of training networks with hidden neurons. Backpropagation computes an error signal for the output neurons and spreads it over the hidden neurons. Contrastive Hebbian learning involves clamping the output neurons at desired values and letting the effect spread through feedback connections over the entire network. To investigate the relationship between these two forms of learning, we consider a special case in which they are identical: a multilayer perceptron with linear output units, to which weak feedback connections have been added. In this case, the change in network state caused by clamping the output neurons turns out to be the same as the error signal spread by backpropagation, except for a scalar prefactor. This suggests that the functionality of backpropagation can be realized alternatively by a Hebbian-type learning algorithm, which is suitable for implementation in biological networks.

Algorithms↗
PubMed2004

Perceptual learning in contrast discrimination and the (minimal) role of context.

Unlike most visual tasks, contrast discrimination has been reported to be unchanged by practice (Dorais & Sagi, 1997; Adini, Sagi, & Tsodyks, 2002), unless practice is undertaken in the presence of flankers (context-enabled learning, Adini et al., 2002). Here we show that under experimental conditions nearly identical to those in the no-flanker practice experiment of Adini et al. (2002), practice significantly improved contrast discrimination. Moreover, in a separate experiment, we found that practice without flankers can improve contrast discrimination to a level only reached with flankers in Adini et al. (2002), but further practice with flankers produces no further improvement of contrast discrimination. These results call into question whether the "context-enabled learning" proposed by Adini et al. (2002) is different from regular contrast learning without flankers. In separate experiments, we found that contrast learning is tuned to spatial frequency, orientation, retinal location, and, unexpectedly, contrast. We also replicated Sagi, Adini, Tsodyks, and Wilkonsky's (2003) more recent finding that no regular contrast learning occurs if reference contrasts are randomly interleaved (contrast roving), and further demonstrated that flankers have no effect on contrast learning under contrast roving, another piece of evidence equating "context-enabled learning" to regular contrast learning. The contrast specificity of learning and the lack of learning under contrast roving provide new evidence in favor of a multiple contrast-selective channels model of contrast discrimination, and against saturating transducer models and multiplicative noise models.

Contrast Sensitivity↗
PubMed2026

Integrating histology and spatial transcriptomics via multimodal transformers and contrastive representation learning for accurate gene expression prediction.

Predicting spatial gene expression from Histological images is a fundamental task in understanding tissue organization and molecular phenotypes. However, existing methods often rely on single-model representations or lack effective alignment between image and transcriptomic features. To address these limitations, we propose a unified multimodal learning framework that integrates histological imaging and spatial transcriptomics through a shared latent representation space. Specifically, histological H&E images are encoded by a ResNet50-based convolutional stem and a MobileViT Transformer backbone to extract hierarchical visual representations. Both modalities are projected into a shared latent space via linear-GELU-dropout transformation blocks, enabling cross-modal alignment through a contrastive learning objective that maximizes agreement between the corresponding image and the spot embeddings. Experimental results on the 10x Genomics Visium dataset of human liver tissue demonstrate that MViTGene achieves significantly higher prediction accuracy than existing methods across multiple gene subsets, with improvements of 20%, 33%, and 12% in predicting marker genes, highly expressed genes, and highly variable genes, respectively. The significant improvement in relevance indicates that the model can more accurately capture the true correspondence between tissue morphology and gene expression, therefore enabling more reliable biological interpretation. It provides a computational tool for high-throughput spatial gene expression prediction that balances performance and interpretability.

Humans↗
PubMed2005

Young children's natural learning environments: contrasting approaches to early childhood intervention indicate differential learning opportunities.

1,000 parents of infants and toddlers enrolled in early childhood intervention programs were surveyed about the number of learning opportunities provided their preschool children using different approaches to early intervention. Findings showed that more learning opportunities were reported when participation in everyday activity settings was conceptualized as a type of intervention rather than as settings for professionals to conduct their interventions.

Child Development↗
PubMed2026

Fairness-aware supervised hierarchical contrastive semantic learning for sexual dimorphism analysis.

MOTIVATION: Sexual dimorphism is a fundamental biological determinant driving systematic differences in disease susceptibility, progression, and clinical outcomes. However, current sex-combined AI-based genomic models often exhibit algorithmic bias and fail to capture these sex-specific mechanisms, creating a critical barrier to unbiased precision medicine. Ensuring fairness in the context of sexual dimorphism requires understanding and addressing the distinct biological mechanisms functioning in each sex, rather than focusing solely on equalizing predictive performance. RESULTS: We propose a fairness-aware supervised hierarchical contrastive learning approach, called FairHICON, to discover unbiased sex-common and sex-specific predictive features. Evaluations on cancer and asthma transcriptomic datasets demonstrate that FairHICON significantly outperforms state-of-the-art benchmarks, improving predictive performance by up to 9% while effectively reducing the performance gap between male and female sexes. Furthermore, prognostic validation confirms that the identified sex-specific pathways stratify patient survival significantly better within their corresponding sex groups. This validates FairHICON to elucidate the molecular heterogeneity of sexual dimorphism, advancing inclusive precision medicine. AVAILABILITY AND IMPLEMENTATION: The source code and data is available at https://github.com/datax-lab/FairHICON.

Sex Characteristics↗