SEARCH · PubMed Health

Results for “multimodal modeling”

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 37 records · Page 2Linked to original sources

PubMed2026

A multi-modal survival prediction framework with group-based batch training and structural consistency alignment.

OBJECTIVE: Integrating whole-slide images (WSIs) with transcriptomic profiles is pivotal for enhancing cancer survival prediction. However, the intrinsic gigapixel resolution and variable sequence lengths of WSIs create a fundamental trade-off between training efficiency and the preservation of data heterogeneity in existing frameworks. Furthermore, substantial statistical and structural discrepancies between histological and genomic modalities often impede effective cross-modal alignment and fusion, thereby limiting prognostic accuracy. METHODS: We propose PRISM, an efficient multi-modal learning framework for integrating WSIs with transcriptomic profiles. To reconcile training efficiency with full data heterogeneity, PRISM first stochastically partitions variable-length WSI sequences into a main subset and a complementary residual subset, both of which are packed into fixed-length groups for batch training. The main subset is processed in the main branch, utilizing isolation masking to maintain intra-group sequence independence. Simultaneously, the residual subset is consolidated into "hyperslides" within a residual branch that leverages tailored supervision, effectively capturing inter-slide correlations. Furthermore, PRISM integrates an Informative Token Aggregation (ITA) module to reduce redundancy in WSIs and employs Cross-batch Structural Consistency Alignment (CBSCA) mechanism to enhance inter-modal structural connectivity. Finally, efficient cross-modal feature interaction is achieved through a Low-rank Bilinear Gated Fusion (LBGF) module. Code is available at https://github.com/Alisa2080/PRISM. RESULTS: Compared with existing methods, PRISM achieves the best overall C-index across five TCGA cohorts. On the larger TCGA-BRCA dataset, PRISM requires only 6 hours of training time, substantially reducing computational cost relative to strong multimodal baselines. Furthermore, comprehensive evaluations demonstrate that PRISM achieves the best overall IBS ranking and favorable time-dependent AUC performance at 1, 3, and 5 years, thereby delivering a more favorable trade-off between prognostic performance and computational efficiency. CONCLUSION: PRISM provides a favorable balance between predictive performance, calibration quality, and computational efficiency, highlighting its potential for practical deployment in multimodal survival modeling for computational pathology.

Humans↗

PubMed2026

Artificial Intelligence for Natural Products Discovery and Development.

Natural products (NPs) remain a cornerstone of modern drug discovery, offering stereochemical complexity and diverse bioactivities that precisely modulate therapeutic targets, refined through billions of years of evolution. However, their research has long been hindered by inefficient, empirical workflows, high resource consumption, structural complexity, and the "multicomponent, multi-target" nature of their mechanisms. The exponential growth of genomic, metabolomic, and spectral data has overwhelmed conventional analytical methods, exposing critical bottlenecks in handling high-dimensional, heterogeneous datasets that exceed human interpretive capacity. Artificial intelligence (AI) is emerging as a transformative paradigm to address these challenges, integrating multi-omics and chemical data to shift NP research from fragmented empiricism toward mechanism-driven, precision-oriented development. By leveraging deep learning architectures- including graph neural networks, Transformers, and diffusion-based generative models-AI enables systematic decoding of NP biosynthesis, automated structure elucidation, rational target identification, knowledge extraction from vast unstructured scientific literature, and de novo molecular design. This review comprehensively surveys recent advances in AI applications across the full NP discovery and development pipeline, encompassing genome mining, structure-based and ligand-based virtual screening, multimodal structural characterization, lead optimization, and biosynthetic pathway engineering. We further examine the emerging roles of protein-centric, molecule- centric, and multimodal foundation models, as well as large language models, in bridging genotype-to-chemotype gaps and unlocking unstructured scientific knowledge. Finally, we discuss critical challenges including data scarcity, representational limitations for complex stereochemistry, physical plausibility in generative models, and the urgent need for experimental validation, while outlining future directions toward autonomous experimentation, closed-loop optimization, and human-AI collaborative discovery.

Artificial intelligence↗

PubMed2026

Joint Effects of Long-Term Obesity and Genetic Susceptibility on Sex-Specific Brain Aging.

OBJECTIVE: This study aimed to examine the associations of longitudinal obesity trajectories and polygenic risk with sex-specific brain aging. METHODS: We analyzed 35,092 UK Biobank participants (16,484 males and 18,608 females). Sex-specific XGBoost models estimated multimodal brain age. We derived 16-year longitudinal obesity trajectories from repeatedly collected anthropometric measurements. Polygenic risk scores were constructed based on 55 independent genetic loci. Multivariable logistic regression examined associations of obesity trajectories and genetic risk with brain age acceleration. RESULTS: A total of 8198 (49.73%) males and 9089 (48.84%) females had accelerated brain aging. High genetic risk significantly increased brain age acceleration odds (males: OR = 1.39; females: OR = 1.34). Crucially, the high-stable obesity trajectory exerted a stronger effect in males (OR = 1.90, 95% CI: 1.64-2.21) than in females (OR = 1.25, 95% CI: 1.12-1.40), with the joint presence of high genetic risk and high-stable obesity amplifying risk to an OR of 2.78 in males and 1.57 in females. Conversely, shifting from obesity to non-obesity reduced risk by 30% in males and 18% in females. CONCLUSIONS: These findings underscore long-term obesity as a critical, sex-dimorphic driver of accelerated brain aging, and midlife weight management offers robust neuroprotection even in genetically susceptible individuals.

brain aging↗

PubMed2026

Machine learning for population-level risk prediction of future cholangiocarcinoma.

BACKGROUND: The poor prognosis of cholangiocarcinoma (CCA) is largely driven by rapid, asymptomatic disease progression, which usually results in a late diagnosis in the absence of established screening strategies. An early, cost-effective, and universally applicable risk assessment strategy would therefore be valuable. METHODS: We developed machine learning (ML) models on prospective, multimodal data from 487,495 UK Biobank (UKB) participants, of whom 649 developed CCA during follow-up. Data from England (80%) were utilised for ML development via five-fold cross-validation, and then all models were tested on withheld data from Scotland, Wales, and Newcastle (20%). Iterative ablation studies reduced inputs from >150 features across demographic data, lifestyle, health records, blood parameters, genomics, and metabolomics to models built on five and ten routinely available clinical parameters. These were externally validated in the Penn Medicine Biobank (PMBB; n = 2638; 28 CCA), All of Us Research Program (AOU; n = 330,433; 362 CCA), Japan Medical Data Centre Claims Database (JMDC; n = 8,425,522; 723 CCA) and TriNetX (n = 728,886; 1592 CCA). FINDINGS: We show that ML models integrating biliary-disease associated health records and Gamma glutamyltransferase can stratify risk of future CCA. Evaluation on the UKB test set as well as three independent cohorts revealed robust performance and generalisability across ethnicities. We achieved AUROCs of 0.71 [95% CI: 0.703-0.711], 0.77 [95% CI: 0.764-0.778 ], 0.796 [95% CI: 0.795-0.798] and 0.8 [95% CI: 0.794-0.805] for UKB, PMBB, AOU, and JMDC respectively, with respective AUPRCs of 0.014 [95% CI: 0.009-0.018], 0.042 [95% CI: 0.037-0.048], 0.038 [95% CI: 0.033-0.042] and 0.001 [95% CI: 0.001-0.001]. In AOU, application of the Youden J-optimised threshold yielded a number needed to screen of 79. Separate models for intra- and extrahepatic CCA did not improve performance. In line with the pathophysiology, performance declined for longer intervals between assessment and event. A group-level analysis in the TriNetX cohort revealed hazard ratios of up to 82.5 [95% CI: 26.4-257.96]. We provide extensive interpretability results and release all source codes used to develop the presented models. INTERPRETATION: We provide a comprehensive framework for early CCA risk stratification in the general population, identifying key predictors, and demonstrating the potential of data-driven models in personalised screening for hepatobiliary cancer. FUNDING: German Cancer Aid (grant #70115730), Junior Principal Investigator Fellowship programme of RWTH Aachen Excellence strategy.

Humans↗

PubMed2026

Biological Foundation Models for Complex Disease Research and Clinical Translation.

Complex diseases, including cancer, rare genetic disorders, neurodevelopmental and psychiatric conditions, and neurodegenerative diseases, arise from interactions among genetic variation, gene regulation, and cellular states that are difficult to capture using a single data type or biological scale. Biological foundation models address this challenge by treating nucleotides and genes as tokens and learning representations that can be transferred to downstream biomedical and clinical tasks. In this review, we examine two major model classes, genomic sequence foundation models and cell foundation models, and compare their tokenization strategies, model architectures, pretraining objectives, and adaptation methods. We summarize their emerging applications in regulatory variant interpretation, disease-associated cell-state analysis, drug-response prediction, and therapeutic target discovery across complex diseases. We distinguish applications supported by experimental or retrospective validation from those that remain primarily computational or conceptual. We further discuss key challenges to clinical translation, including multimodal data integration, model interpretability, benchmarking, patient-specific prediction, and privacy protection. We highlight future opportunities to integrate biological foundation models with emerging frameworks of medical digital twins, agentic AI, and federated learning. By linking model design to translational goals, this review provides a practical framework for evaluating biological foundation models and their readiness for complex disease research and clinical use.

biological foundation model↗

PubMed2026

Harnessing the Power of Large Language Models for Drug Discovery: A Systematic Review of Current Applications and Future Directions.

INTRODUCTION: The demand for inventive approaches to drug discovery has increased due to the rising costs, time, and failure rates in pharmaceutical research. Large Language Models (LLMs), with their sophisticated natural language processing and generative capabilities, have become potent instruments that have the potential to revolutionize biomedical research. The function of LLMs in different phases of drug development is methodically examined in this article. METHODS: The PRISMA 2020 principles were adhered to in this systematic study. A thorough search for research published between 2018 and 2025 was done using PubMed, Scopus, Web of Science, and Google Scholar. The search terms "large language model," "transformer," "drug discovery," and important sub-domains (such as "de-novo design" and "ADMET") were merged, and two reviewers independently screened the results. Predetermined inclusion and exclusion criteria were used to filter studies for relevance. 98 studies out of the 1,285 records that were initially retrieved met the requirements for the final qualitative synthesis. RESULTS: 98 studies that demonstrated the use of LLMs in various drug discovery domains were found during the review. These covered molecular generation, genomics, protein-ligand modeling, ADME/T and toxicity profiling, drug-target interaction and DTI prediction, and biomedical text mining. 42 different LLM-based tools were mapped, including BioBERT, SciSpacy, Drug- LLM, DNA-BERT, GPT-4, and ChatGPT. Predictive accuracy, hypothesis creation, target prioritization, and multi-modal data integration all showed notable gains with these techniques. DISCUSSION: By providing scalable, precise, and effective solutions for data-driven drug discovery, LLMs are revolutionizing the pharmaceutical industry. They allow for the creation of hypotheses and individualized insights across multi-modal biological data, and they perform better than conventional approaches in a number of subdomains. Improvements in performance were task-dependent; the most consistent gains occurred for biomedical text mining, disease-genedrug relationship mapping and drug-target interaction prediction tasks. Yet most evidence for clinical applications is still derived from retrospective studies and benchmark datasets, suggesting a higher need for prospective validation. CONCLUSION: There is revolutionary potential in incorporating LLMs into drug discovery processes. Clinical translation and regulatory uptake will depend heavily on collaborative validation, ethical deployment, and standardization as models become more multimodal and interpretable. Before normal use, extensive prospective benchmarking and head-to-head comparisons with established chemoinformatics pipelines are necessary.

De novo design↗

PubMed2026

Large language models in bioinformatics: a comprehensive survey.

The emergence of foundation models with trillion-level parameters has redefined the landscape of artificial intelligence. Various fields are developing their own large-scale models, which can solve many problems within the field and improve work efficiency. Biological large-scale models are a cross-disciplinary research field that combines mathematics, computer science, and biology, aiming to simulate and understand the structure, function, and dynamic changes of biological systems through the establishment of complex computational models. This field covers multiple levels such as biological pathways, population dynamics, protein folding, etc., providing us with tools for deep exploration of the mysteries of life and applications in medicine, ecology, and other fields. This article reviews the background and research status of biological large-scale models, and discusses future directions. Large language models (LLMs) and other large-scale foundation models have rapidly advanced in recent years, enabling powerful representation learning and generation across text, sequences, and multimodal data. In bioinformatics and biomedicine, these models are increasingly used to analyze genomic sequences, infer protein properties and structures, support drug discovery, and integrate heterogeneous biomedical evidence. This survey reviews the basic principles of LLMs and summarizes representative applications in (i) gene and genome sequence analysis, (ii) protein structure and function prediction, and (iii) drug design, including virtual screening and personalized medicine. We also discuss emerging multi-model modeling approaches, as well as key challenges such as data quality and privacy, interpretability, generalization to new organisms and tasks, and responsible deployment in health-related settings. Finally, we outline future directions for developing reliable, scalable, and explainable bioinformatics foundation models.

bioinformatics↗

PubMed2026

Dual-Reporter Gene-Based Multimodal Imaging for Tracking Mesenchymal Stem Cells in Diabetic Skin Wound Repair.

BACKGROUND: Diabetic foot ulcer (DFU) is a clinically challenging complication characterized by poor healing outcomes, and conventional therapies provide limited benefit. Mesenchymal stem cell (MSC) transplantation offers a promising strategy for DFU repair. However, the low survival of transplanted MSCs in the hostile wound microenvironment, coupled with the lack of real-time, non-invasive methods to track these cells in vivo, severely hampers their therapeutic efficacy and clinical translation. METHODS: We engineered MSCs to co-express a dual reporter system comprising near-infrared fluorescent protein (iRFP) and ferritin heavy chain (FTH1). These modified cells were then integrated with a fibrin glue (FG) scaffold to create a unified platform that supports both multimodal imaging and therapeutic function within skin wounds. First, FTH1 overexpression enhances the antioxidant capacity of MSCs, while the FG scaffold provides structural support; this combination enhances cell survival and retention. Second, the iRFP/FTH1 dual reporter enables near-infrared fluorescence imaging and MRI-based localization, establishing a multimodal platform for real-time cell tracking. RESULTS: In a full-thickness skin defect model in diabetic mice, multimodal imaging revealed that transplanted cells persisted in the wound area for approximately seven days. Treatment with iRFP/FTH1-MSCs/FG significantly accelerated wound closure and promoted hair follicle regeneration and angiogenesis. Additionally, local iron deposition resulting from FTH1 expression enhanced fibroblast migration and collagen synthesis, further facilitating extracellular matrix remodeling. Mechanistic studies demonstrated that this therapy drives macrophage polarization toward the anti-inflammatory M2 phenotype and activates the PI3K-AKT-VEGF signaling pathway. These complementary effects synergistically enhance tissue regeneration and systematically improve diabetic wound healing. CONCLUSIONS: Collectively, this multimodal stem cell-scaffold system effectively integrates dynamic cell tracking with stem cell therapy during skin wound repair. It addresses a critical technical gap in visualizing stem cells within the wound microenvironment and provides valuable methodological and theoretical foundations for optimizing regenerative strategies for diabetic skin wounds.

Animals↗

PubMed2026

Deep learning-based multimodal pathogenomics integration for precision cancer prognosis.

BACKGROUND: Recent studies have revealed valuable prognostic insights in haematoxylin and eosin (H&E)-stained histological sections and transcriptomic profiles, suggesting potential applications in machine learning. However, existing methods lack sufficient intra- and inter-modal interactions, and face challenges in clinical validation due to incomplete multimodal data. METHODS: We proposed PathoGems (PathoGenomics-based integrative survival prediction), a weakly-supervised, interpretable multimodal learning framework that integrates histology and genomic profiles for precise cancer prognosis prediction. To evaluate the robustness of PathoGems, we initially curated a dataset of 1965 cases across four cohorts from The Cancer Genome Atlas (TCGA), including breast, colorectal, glioblastoma, and esophageal cancers. For external validation, PathoGems was further evaluated on four independent cohorts, consisting of 76 breast cancer and 41 esophageal squamous cell carcinoma cases from Zhejiang Cancer Hospital, as well as 102 colorectal cancer and 58 glioblastoma cases from the Clinical Proteomic Tumor Analysis Consortium (CPTAC). RESULTS: PathoGems effectively stratified patients into favorable and unfavorable risk groups, revealing significant differences in histological patterns, genomic features, and overall survival (log-rank test, p < 0.05). Moreover, the model’s predictions are further supported by visualization and transcriptomic analysis, enhancing interpretability and reliability. CONCLUSIONS: By fusing histological and clinicogenomic multimodal models, PathoGems will provide a solid foundation for developing an innovative tool that aids clinicians in making informed decisions and selection personalized treatment strategies for cancer patients.

Humans↗

PubMed2026

Multimodal alignment improves generalizability of genomic biomarker prediction in computational pathology.

Computational pathology models that use digitized histopathology whole-slide images have the potential to become a cost-effective and scalable alternative to molecular assays for the prediction of genomic biomarkers, a key task in precision oncology. However, as new genomic biomarkers are discovered or quantified, large, labeled datasets must be prospectively collected to train new models. To address this challenge, we developed multimodal alignment for biomarker learning and generalization (MARBLE), a multimodal contrastive pretraining strategy that integrates structured biomarker knowledge into representation learning of histopathology images. MARBLE aligns histopathology-derived representations with representations of genomic biomarkers generated by a large language model (LLM) and a protein language model (PLM). This biologically informed alignment enables data-efficient generalization to novel, out-of-distribution biomarkers. Using the MSK-IMPACT cohort of over 40,000 patients across multiple biomarker panel versions, we design experiments grounded in real-world data to demonstrate the value of our proposed approach.

CP: computational biology↗

PubMed2026

AI-Supported, Integrative Prediction of Postoperative Delirium: Protocol for the CONFUSED Study.

BACKGROUND: Postoperative delirium (POD) is a frequent and serious complication in older surgical patients, characterized by acute cognitive dysfunction and fluctuating levels of consciousness. POD is associated with prolonged hospitalization, long-term cognitive decline, reduced quality of life, and increased mortality. Despite its clinical relevance, the underlying pathophysiological mechanisms remain poorly understood, and reliable biomarkers for early prediction and prevention are lacking. OBJECTIVE: The CONFUSED study aims to identify molecular and clinical predictors of POD by integrating clinical data with proteomic, transcriptomic, and epigenetic analyses. The primary objective is to develop predictive models for POD using multimodal data. Secondary objectives include the identification of delirium-associated genes, proteins, and epigenetic signatures, as well as the exploration of patient subgroups at increased risk for POD. METHODS: CONFUSED is a prospective observational cohort study conducted at a German university hospital. Adult patients undergoing major surgery under general anesthesia will be enrolled until 100 cases of POD have been observed, which is expected to require a total sample size of approximately 200 to 300 patients. Blood samples are collected at 4 predefined time points: before premedication, immediately after surgery, and on postoperative days 2 and 5. Samples undergo comprehensive proteomic profiling, transcriptomic analysis using RNA microarrays, DNA methylation analysis, and genotyping of selected polymorphisms. Clinical data, including demographics, comorbidities, perioperative variables, medications, and delirium assessments using the Confusion Assessment Method (CAM) and CAM for the intensive care unit, are systematically recorded. Statistical analyses include univariate and multivariate methods, as well as machine learning approaches such as random forests and support vector machines, to identify relevant biomarkers and develop predictive models. The study protocol follows STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) and TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) guidelines and was approved by the responsible ethics committees. RESULTS: The study was registered in the German Clinical Trials Register (DRKS00033854) on March 18, 2024. Recruitment started in January 2024 and is ongoing at the time of manuscript submission. As of now, 135 patients have been enrolled. Sample collection and laboratory analyses are ongoing. Data analysis began in January 2026, with first results anticipated in July 2026. Final data lock is anticipated after the completion of recruitment. CONCLUSIONS: By integrating multimodal molecular data with clinical parameters and applying advanced machine learning techniques, the CONFUSED study aims to improve the prediction and understanding of POD. The results are expected to support the development of personalized preventive strategies and contribute to improved perioperative care for patients at risk of POD.

Humans↗

PubMed2026

H&E to recurrence score: A step forward, but not yet a substitute for genomic testing.

Shamai and colleagues developed a multimodal deep-learning model that predicts Oncotype DX recurrence scores from routine H&E slides and clinicopathological variables in hormone receptor‑positive, HER2‑negative early breast cancer. Validated across the TAILORx trial and six external cohorts (over 5000 patients), the model achieved an AUC of 0.898 for identifying recurrence score ≥26 and recapitulated genomic assay patterns of chemotherapy benefit. Notably, 31% of clinically high-risk postmenopausal women were downgraded to low risk by AI, suggesting potential to reduce overtreatment. However, several limitations preclude immediate clinical substitution for genomic testing. First, intratumoural heterogeneity leads to discordant predictions with unclear management guidance. Second, the model's chemotherapy benefit estimates rely on TAILORx's age-based menopausal surrogates, which may not reflect real-world hormonal status or LHRH agonist use. Third, predictive value in node-positive disease remains untested in randomised datasets such as RxPONDER. Additionally, calibration uncertainty near risk thresholds and global scalability issues (including IHC requirements and digital pathology infrastructure) persist. While this represents a landmark step toward democratising precision oncology, the AI tool should currently serve as a complementary decision aid, with genomic testing remaining the gold standard for intermediate, borderline, or discordant cases.

Breast cancer↗

PubMed2025

Multimodal profiling reveals tissue-directed signatures of human immune cells altered with age.

The immune system comprises multiple cell lineages and subsets maintained in tissues throughout the lifespan, with unknown effects of tissue and age on immune cell function. Here we comprehensively profiled RNA and surface protein expression of over 1.25 million immune cells from blood and lymphoid and mucosal tissues from 24 organ donors aged 20-75 years. We annotated major lineages (T cells, B cells, innate lymphoid cells and myeloid cells) and corresponding subsets using a multimodal classifier and probabilistic modeling for comparison across tissue sites and age. We identified dominant site-specific effects on immune cell composition and function across lineages; age-associated effects were manifested by site and lineage for macrophages in mucosal sites, B cells in lymphoid organs, and circulating T cells and natural killer cells across blood and tissues. Our results reveal tissue-specific signatures of immune homeostasis throughout the body, from which to define immune pathologies across the human lifespan.

Humans↗

PubMed2026

Deep generative models in biological sequence and structure analysis and design.

Deep generative models have transformed biological sequence modeling from predictive analysis toward increasingly controllable design. Early biological applications of Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) established latent representation learning and sequence synthesis, while recent advances in transformer-based language models, discrete diffusion, flow-matching, and multimodal generative frameworks have substantially expanded the scope of biological design. This review examines generative models for DNA, RNA, and protein sequence design, emphasizing how different model classes represent biological constraints, operate over discrete and continuous spaces, and integrate sequence, structure, and function. We compare VAEs, GANs, autoregressive and masked language models, diffusion models, and flow-based approaches across genomics, transcriptomics, and proteomics, with particular attention to controllability, long-range dependency modeling, structural grounding, generalization, and experimental utility. We further examine evaluation strategies, out-of-distribution generalization, and closed-loop design-build-test-learn workflows that connect in silico generation with empirical validation. We distinguish fundamental modality-dependent constraints including sequence discreteness, context length, structural coupling, and physical or thermodynamic requirements from architecture-dependent advantages that reflect the current state of the field. Current studies suggest that long-context models are particularly useful for genome-scale representation and sequence modeling, whereas structure-aware diffusion, flow-based, and inverse-folding approaches provide better frameworks for geometry-constrained RNA and protein design. This perspective provides a critical framework for understanding the present capabilities, limitations, and convergence of generative approaches toward reliable and experimentally grounded biological design.

Biological sequence analysis↗

PubMed2026

CodonMoE: DNA language models for codon-dependent mRNA prediction.

MOTIVATION: Genomic language models (gLMs) face a fundamental efficiency challenge: one must either maintain separate specialized models for each biological modality (DNA and RNA) or develop large multimodal architectures. Both approaches impose significant computational burdens-modality-specific models require redundant infrastructure despite inherent biological connections, while multi-modal architectures demand increased parameter counts and extensive cross-modality pretraining. RESULTS: To address this limitation, we introduce CodonMoE (Adaptive Mixture of Codon Reformative Experts), a lightweight adapter that transforms DNA language models into effective RNA analyzers without RNA-specific pretraining. Our theoretical analysis establishes CodonMoE as a universal approximator at the codon level, capable of mapping arbitrary functions from codon sequences to codon-dependent RNA properties given sufficient expert capacity. Across four RNA prediction tasks spanning stability, expression, and regulation, DNA models augmented with CodonMoE significantly outperform their unmodified counterparts, with the HyenaDNA+CodonMoE series achieving state-of-the-art results using 80% fewer parameters than specialized RNA models. By maintaining sub-quadratic complexity while achieving superior performance, our approach provides a principled path toward unifying genomic language modeling, leveraging more abundant DNA data and reducing computational overhead while preserving modality-specific performance advantages. AVAILABILITY AND IMPLEMENTATION: Source code for the method and to reproduce the results is available at https://github.com/Kingsford-Group/CodonMoE.

Codon↗

PubMed2026

BriGHT: transcriptome-regularized multimodal neuroimaging for brain disorder prediction.

MOTIVATION: Hypergraph-based models for brain disorder prediction mainly adopt imaging-derived hypergraphs as propagation backbones. However, the entanglement of topology construction and feature propagation leaves regional representations weakly constrained by underlying biological organization, making them vulnerable to subject-specific variation and noise, particularly in heterogeneous multimodal settings. RESULTS: We present BriGHT, a Brain transcriptome-reGularized Hypergraph framework for mulTimodal disorder prediction. BriGHT employs a transcriptome-derived structural reference as a soft anchoring prior to regularize neuroimaging ROI embeddings, stabilizing representation geometry while preserving disease-relevant subject-specific variation. BriGHT further incorporates a reliability-aware fusion module to estimate subject-specific modality reliability from prediction confidence, cross-modal consistency, and decision certainty, enabling adaptive integration under heterogeneous modality quality. Experiments on three neuroimaging cohorts (ADNI, ADHD-200, REST-meta-MDD) and four modalities (VBM, fMRI, FDG, AV45) demonstrate that BriGHT consistently outperforms competing graph/hypergraph learning methods across six brain disorder prediction tasks. Perturbation analyses show that BriGHT benefits from the spatial correspondence between transcriptomic modules and imaging ROIs, rather than from arbitrary hypergraph regularization alone. Ablation and meta-analytic interpretability analyses support the contribution of transcriptomic anchoring and adaptive fusion to robust and biologically meaningful brain disorder prediction. AVAILABILITY: The software is publicly available at: https://github.com/Yaolab-fantastic/BriGHT. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Journal Article↗

PubMed2026

Automated CEAP Classification of Venous Duplex Reports Using Multimodal Artificial Intelligence.

OBJECTIVE: To develop and internally validate a prototype multimodal artificial intelligence system for automated CEAP (Clinical, Etiological, Anatomical and Pathophysiological) classification of venous duplex ultrasound (VDUS) reports, integrating natural language processing of free-text components with computer vision analysis of hand-drawn anatomical diagrams. METHODS: Single centre retrospective observational study using routinely collected clinical data. One thousand consecutive venous duplex ultrasound reports from Cambridge University Hospitals NHS Foundation Trust, UK (July 2024 - May 2025) were labelled according to the CEAP classification, excluding the Etiological component, which could not be reliably determined from duplex reports alone. Transfer learning was applied using ClinicalBERT for text and MobileNetV3 for diagrammatic data. Clinical classes were predicted from request line text. Text- and image-based pathophysiological models were developed for four anatomical territories (Great Saphenous Vein, Small Saphenous Vein, Deep system, Perforators), combined using late fusion with probability averaging. RESULTS: The clinical CEAP model achieved accuracy of 0.91, macro-F1 of 0.82, and macro-AUC of 0.98. Pathophysiological prediction varied, with text models broadly outperforming image models. Fusion yielded heterogeneous benefits, improving SSV performance but reducing Deep system accuracy. The performance of the final pathophysiological CEAP fusion models varied across anatomical territories: accuracy ranged from 0.70-0.92 and macro-AUC from 0.80-0.92. CONCLUSION: This study demonstrates the feasibility of automated CEAP classification from VDUS reports. Despite class imbalance affecting minority class predictions, the strong discriminatory performance validates this multimodal ML model for extracting clinically meaningful information from real-world data. This approach offers potential, pending external validation, to streamline vascular services through automated triage and guideline-compliant decision making.

Artificial intelligence↗

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↗