Zixin Hu

Organ-level investigations provide insights into inter-individual heterogeneity in health and disease, yet aging across organs and systems remains incompletely understood. Leveraging plasma proteomic data from 53,014 UK Biobank participants, we construct 16 organ- and system-specific aging clocks that show 5436 significant associations with 1059 diseases (i.e., 655 incident and 404 prevalent diseases) and widespread responses to 841 diverse environmental factors, highlighting organ-specific and pan-organ patterns. These clocks are benchmarked against two existing models and then externally validated in the China Kadoorie Biobank (n = 3977), supporting their robustness and generalizability. Genetic analyses identify 261 genomic loci associated with these clocks, revealing organ-specific and pleiotropic genetic architectures. Notably, the immune system and liver exhibit the strongest genetic connectivity with other organ systems, closely mirroring their phenotypic correlations. Using longitudinal data from 1006 participants, we apply a multitask Bayesian framework to explore potential temporal interaction networks across organs and systems, identifying temporal precedence of the immune system, liver and stomach, with sex-specific patterns characterized by stronger arterial precedence in females and stomach precedence in males. Together, this study provides a comprehensive atlas integrating genetic, environmental, and interactive insights into multiple plasma protein-based organ and system aging, informing future research on targeted anti-aging strategies.

Background: The impact of dietary fat on dementia risk remains controversial due to conflicting findings on fat subtypes, over-reliance on linear models, and limited attention to fat source balance. Methods: In a large prospective analysis of 97,765 UK Biobank adults, we examined associations between dietary fat composition and incident dementia, integrating genetic susceptibility and metabolomic profiling. Results: Total fat intake, animal fat, and plant fat all showed significant U-shaped associations with dementia risk. Notably, the animal-to-plant fat ratio exhibited a clear U-shaped association ( P _FDR for nonlinearity = 0.011), with the lowest risk observed at a ratio of 0.9. A significant interaction with APOE -ε4 carriage was identified ( P for interaction = 0.041), in which carriers in the lowest quartile of the fat ratio had the highest dementia risk (HR = 3.88, 95% CI: 3.00–5.00). A similar nonlinear pattern was observed for executive function. Mediation analyses implicated very-low-density lipoprotein-related metabolites as key intermediates. Conclusions: These findings highlight the importance of balanced fat sources and support precision nutrition strategies tailored to genetic risk.

Large language models (LLMs) have been rapidly adopted in healthcare since 2022; however, field-level trends and specialty differences remain poorly characterized. This study aims to map the research landscape of LLMs in healthcare and generate comparative specialty profiles through a bibliometric analysis using the Web of Science Core Collection (2015–2025), including English-language articles and reviews analyzed for publication and citation trends, leading countries, institutions, authors, journals, co-citation networks, and keyword structures with VOSviewer and CiteSpace, with sub-analyses for medicine, general and internal; surgery; and radiology, nuclear medicine and medical imaging. Of the 2226 articles included, the annual output increased from four publications in 2022 to 1327 in 2025, yielding a compound annual growth rate (CAGR) of 592.26%; the United States of America contributed 43.5% of the publications, followed by China (14.3%), Germany (9.2%), Turkey (9.0%), and England (6.8%); the publications for each specialty were surgery (31.7%), medicine, general and internal (24.2%), and radiology, nuclear medicine and medical Iimaging (13.0%), with the last demonstrating the fastest 2023–2024 growth (CAGR 265%); and the collaboration networks were United States of America-centered, with dense trans-Atlantic ties. LLM research in healthcare is expanding rapidly with distinct specialty-specific trajectories, and interpreting these trajectories using the DECIDE-AI (developmental and exploratory clinical Iinvestigations of decision support systems driven by artificial intelligence) framework clarifies low-risk near-term applications and monitoring priorities, providing a specialty-aware baseline to support the safe, equitable, and regulation-aligned adoption of LLMs in clinical practice.

Massive rotator cuff tears (MRCT) remain a clinical challenge, characterized by poor tendon‐bone interface (TBI) healing, severe muscle degeneration, and high postoperative retear rates. Tissue engineering scaffolds offer promising alternatives, yet traditional biomaterials often lack sufficient bioactivity to orchestrate comprehensive tissue regeneration. Herein, we developed a novel graphene oxide (GO)‐engineered porcine type I collagen (GO/Col) scaffold and systematically investigated its therapeutic efficacy and underlying molecular mechanisms via multi‐omics analyses. Comprehensive characterization showed that GO incorporation improved scaffold stability, wettability, and biocompatibility. In vitro, the GO/Col scaffold enhanced mesenchymal stem cell adhesion and proliferation, promoted osteogenic and chondrogenic differentiation, and suppressed adipogenesis. In a macrophage model system, GO/Col was associated with a shift toward a more reparative, anti‐inflammatory phenotype. Using a clinically relevant chronic MRCT rat model, we observed that GO/Col scaffolds significantly improved motor function, biomechanical properties, and tendon‐bone regeneration, while inhibiting muscle fibrosis and fatty infiltration. Mechanistically, integrated transcriptomics, proteomics, and mass cytometry analyses revealed GO‐mediated modulation of critical signaling pathways involved in immune regulation, stem cell differentiation, and tissue regeneration. Notably, GO activated pro‐osteogenic/chondrogenic pathways and anti‐inflammatory signatures, while downregulating adipogenic and pro‐inflammatory pathways. Collectively, these findings support the potential of GO/Col scaffolds as a bioactive tissue‐engineering strategy for chronic MRCT repair.

Despite the widespread success of Graph Neural Networks (GNNs), understanding the reasons behind their specific predictions remains challenging. Existing explainability methods face a trade-off that gradient-based approaches are computationally efficient but often ignore structural interactions, while game-theoretic techniques capture interactions at the cost of high computational overhead and potential deviation from the model's true reasoning path. To address this gap, we propose FSX (Message Flow Sensitivity Enhanced Structural Explainer), a novel hybrid framework that synergistically combines the internal message flows of the model with a cooperative game approach applied to the external graph data. FSX first identifies critical message flows via a novel flow-sensitivity analysis: during a single forward pass, it simulates localized node perturbations and measures the resulting changes in message flow intensities. These sensitivity-ranked flows are then projected onto the input graph to define compact, semantically meaningful subgraphs. Within each subgraph, a flow-aware cooperative game is conducted, where node contributions are evaluated fairly through a Shapley-like value that incorporates both node-feature importance and their roles in sustaining or destabilizing the identified critical flows. Extensive evaluation across multiple datasets and GNN architectures demonstrates that FSX achieves superior explanation fidelity with significantly reduced runtime, while providing unprecedented insights into the structural logic underlying model predictions--specifically, how important sub-structures exert influence by governing the stability of key internal computational pathways.

Background: In intensive care units (ICUs), managing septic shock requires maintaining adequate tissue perfusion with vasopressors, most commonly norepinephrine, while avoiding under or over-dosing that can worsen hypotension, organ injury, and adverse effects. Bedside vasopressor titration often depends on clinician judgment and simple rules, with limited tools providing individualized, time-aware guidance or early warning of impending hypotension. ChronoSynthNet aimed to create a data-driven model that learns from routine electronic health record (EHR) time-series data to personalize vasopressor therapy and anticipate deterioration. To develop and validate a dual-task deep learning model that predicts real-time norepinephrine requirements and detects hypotension early in adults with septic shock. Methods: We performed a retrospective cohort analysis using the Medical Information Mart for Intensive Care [MIMIC-IV (2008–2019)] database. Eligible adult ICU stays met Sepsis-3 criteria, received norepinephrine, and had adequate time-series data. ChronoSynthNet integrates a shared Transformer encoder, long short-term memory (LSTM) layers, and a dynamic feature-weighting network to learn cross-variable and temporal relationships. The dataset was split 80/20 into training and internal test sets, with five-fold cross-validation on training data. Classification performance for early hypotension detection was assessed using area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), precision, recall, and specificity; norepinephrine rate prediction performance was assessed using mean squared error (MSE). Ninety-five percent confidence intervals (95% CIs) were calculated for AUROC, recall, and specificity on the internal test set using bootstrap and Wilson methods. Results: ChronoSynthNet achieved AUROC of 0.89 (95% CI: 0.836–0.938) for hypotension classification and MSE of 0.0213 (95% CI: 0.0192–0.0234) for predicting the norepinephrine infusion rate. The model demonstrated high specificity (97%, 95% CI: 96.3–98.3%) and precision (92%, 95% CI: 90.3–93.7%), with a recall of 74% (95% CI: 71.3–76.7%). Hypotension events were predicted a median of 3.5 hours in advance. Conclusions: ChronoSynthNet demonstrated strong performance in early hypotension detection and norepinephrine dose forecasting in ICU patients with septic shock. These findings support its potential role in aiding real-time vasopressor titration and early recognition of hemodynamic instability; prospective multicenter validation is needed before clinical deployment.

Biological age reflects actual ageing and overall health, but current ageing clocks are often complex and difficult to interpret, which limits their clinical application. This study introduces a Gompertz law‐based biological age (GOLD BioAge) model designed to simplify the assessment of ageing. We calculated GOLD BioAge using clinical biomarkers and found significant associations between the difference from chronological age (BioAgeDiff) and the risks of morbidity and mortality in the NHANES and UK Biobank. Using proteomics and metabolomics data, we developed GOLD ProtAge and MetAge, which outperformed the clinical biomarker models in predicting mortality and chronic disease risk in UK Biobank. Benchmark analyses demonstrated that the models outperformed common ageing clocks in predicting mortality across diverse age groups in both the NHANES and UK Biobank cohorts. Additionally, a simplified version called Light BioAge is created, which uses three biomarkers to assess ageing. The Light model reliably captured the mortality risk across three validation cohorts (CHARLS, RuLAS, and CLHLS). It significantly predicted the onset of frailty, stratified frail individuals, and collectively identified individuals at high risk of mortality. In summary, the GOLD BioAge algorithm provides a valuable framework for the assessment of ageing in public health and clinical practice.

Proteomics enables systematic elucidation of the biological mechanisms underlying health states including frailty. Here, through a large-scale proteome-wide association study (PWAS) encompassing 2,911 plasma proteins in 50,506 UK Biobank participants, we identified 1,339 proteins significantly associated with frailty, revealing novel functional modules implicated in frailty pathogenesis, particularly the one characterized by the collagen-containing extracellular matrix and vesicle lumen pathways. Replication analyses in an independent external cohort (TwinGene study) confirmed partial but consistent associations at both protein and pathway levels, supporting the reliability of these findings. Mendelian randomization analyses supported causal associations of 50 proteins with frailty. Protein-protein interaction network and expression quantitative trait loci analyses revealed MMP1 and LGALS8 serving as hub proteins. Moreover, we developed a novel proteome-based frailty measure, termed as Proteomic Frailty Score (PFS), which demonstrated robust predictive performance (C-index > 0.7) for 198 (30.2% = 198/655) incident diseases across 13 categories and broad responsiveness to 85 modifiable risk factors. Incorporating PFS into a conventional risk factors model significantly improved the predictive performance for 510 (77.9% = 510/655) incident diseases. Longitudinal analyses with three assessments (n∼1000) revealed an accelerated progression of the PFS with advancing age and increasing baseline frailty severity. To facilitate public use, we further created a publicly accessible online tool for PFS calculation ( https://zipoa.shinyapps.io/frailty/ ). Finally, we observed a biphasic pattern of frailty-associated proteomic dysregulation across lifespan, with peak transitions occurring at approximately ages 50 and 63, implicating distinct biological pathways. Together, we establish PFS as a robust biomarker of biological aging while identifying critical windows and molecular targets for interventions against frailty progression.