High-throughput zebrafish screening generates large volumes of imaging data, creating a need for scalable computational methods that reduce dependence on manual annotation and repeated task-specific model development. In this study, we develop a self-supervised foundation model for zebrafish image analysis pretrained on more than 468,000 images from seven heterogeneous zebrafish imaging datasets spanning different developmental stages, imaging platforms, modalities, and experimental conditions. By learning a shared representation from diverse unlabeled zebrafish images, the proposed framework provides a reusable encoder that can be adapted to multiple biological endpoints and annotation types, reducing the need to learn independent representations for each downstream task. We evaluated two self-supervised pretraining strategies: Self-Distillation with No Labels (DINO) and a sequential Masked Autoencoder (MAE) followed by DINO (MAE+DINO), across morphology classification, view classification, developmental stage prediction, fish localization, embryo clustering, and anatomical segmentation. The learned representations captured visual similarity, developmental progression, embryo-specific trajectories, phenotypic information, and anatomical structure across heterogeneous imaging data. MAE+DINO achieved the strongest overall performance for morphology classification, view classification, developmental stage prediction, and fine-tuned localization, while DINO provided the strongest frozen representation for localization. For anatomical segmentation, zebrafish specific pretraining improved performance when the encoder was frozen, whereas MAE+DINO and ImageNet initialization achieved comparable overall performance after end-to-end fine-tuning with a SegFormer-style decoder. These results demonstrate that organism-centered self-supervised pretraining can provide reusable and biologically informative visual representations that support diverse zebrafish image-analysis tasks and offer a scalable, annotation-efficient alternative to repeatedly developing task-specific representations.
The growing recognition of the human microbiome as a key regulator of immune homeostasis has accelerated the application of computational intelligence for microbiome-driven disease understanding and therapeutic design. However, existing microbiome studies largely rely on classical machine learning or shallow deep learning models that fail to capture higher-order microbial interactions, multimodal functional dependencies, and immune feasibility constraints simultaneously. Moreover, most approaches lack biological constraint enforcement, leading to predictions that may be statistically accurate but immunologically implausible. To address these limitations, this study introduces SIMT, the Synthetic Immune Modulation Transformer, a novel immune-aware deep learning framework for microbial interaction modelling and synthetic microbiota design. SIMT integrates a graph transformer for microbe–microbe interaction learning, a multimodal transformer for immune-associated functional inference, and a newly proposed Immune-Aware Constraint Layer (IACL) that enforces immune feasibility and homeostasis during optimization. The framework operates by learning weighted microbial interaction networks, integrating taxonomic abundance with inferred functional pathways, and constraining latent representations to physiologically meaningful immune ranges. The entire pipeline was implemented using Python-based deep learning libraries for scalable and reproducible analysis. Experimental evaluation demonstrated that the proposed approach achieved an F1-score of 96.41% and an AUC of 95.12%, outperforming existing microbiome-based models, including Random Forest, explainable RF frameworks, convolutional neural networks, fine-tuned language models, and regularized logistic regression reported in prior studies. Beyond predictive performance, SIMT enables immune-stable synthetic consortium optimization, offering interpretable and biologically grounded insights. Overall, the results confirm that immune-aware transformer modelling significantly advances microbiome analytics, supporting reliable in silico design of immune-compatible microbial communities for translational biomedical applications.
PURPOSE:This study aimed to identify the signature genes that mediate the effects of type 2 diabetes (T2D) on coronary artery bypass grafting (CABG), and to elucidate their molecular regulatory mechanisms and potential clinical therapeutic value for improving the clinical outcomes of T2D patients undergoing CABG. METHODS:An integrated approach of bioinformatics analysis, machine learning, clinical validation and in vitro experiments was employed. Multi-cohort T2D and CABG data were analyzed via differential expression profiling, functional enrichment, PPI network construction, and LASSO regression/Boruta algorithms to screen core signature genes mediating the T2D-CABG interaction. Nomogram models were built and validated to assess the diagnostic value of these genes. LncRNA-miRNA-mRNA regulatory networks were constructed to explore their molecular mechanisms. Potential targeted drugs were predicted via DSigDB mining and molecular docking. Clinical validation was performed in CABG patients, and human aortic smooth muscle cells were used for in vitro experiments to verify the functional role of the key signature gene SLCO5A1. RESULTS:MINPP1, HES4 and SLCO5A1 were identified as core signature genes mediating the adverse effects of T2D on CABG, with consistent abnormal expression in T2D and CABG datasets and all AUC values > 0.625. Nomogram models based on these genes showed excellent calibration and clinical net benefit for evaluating T2D-associated pathological risks in CABG patients. JUN and GATA2 were found to be key transcription factors regulating these genes, and a novel complex lncRNA-miRNA-mRNA regulatory network was constructed. Clinical detection revealed significantly up-regulated SLCO5A1 in T2D patients undergoing CABG, with its expression positively correlated with systemic inflammatory burden. In vitro experiments confirmed that SLCO5A1 knockdown markedly inhibited T2D-related metabolic stress (high glucose/Ox-LDL)-induced vascular inflammation, endothelial dysfunction, oxidative stress and cytoskeletal remodeling via the NF-κB pathway. Molecular docking demonstrated high binding affinity of CHEMBL1182312 to SLCO5A1 (docking score = -6.9 kcal/mol) and stable binding of valproic acid to MINPP1 (docking score = -5.0 kcal/mol). CONCLUSION:MINPP1, HES4 and SLCO5A1 were signature genes mediating the effects of T2D on CABG.
Nanofiber production technology has always been interesting as there are a variety of flexible ways to produce one-dimensional organic, inorganic, and constructed nanomaterials with controllable dimensions. Electrospinning is a simple and affordable method for the production of nanofibers, providing large specific surfaces and highly porous structures with diameters ranging from nanometers to micrometers. This process is based on electrostatic fields and precisely controls the dimensions and morphology of the fibers through parameter optimization and the use of special spinning and collectors. This paper covers Electrospinning processes and parameters in detail and illuminates the factors that influence Electrospinning. It deals with the morphological and structural aspects of Electrospinning fibers used in different applications. Additionally, this paper examines the various polymers and non-polymeric materials used in Electrospinning, and the properties and applications of nanofibers produced by the electrospinning method are investigated. Nanofibers have a variety of applications, including defense industries, tissue engineering, filtration, wearable biosensors, cosmetics, etc. Additionally, we will examine the inclusion of fillers in the Electrospinning to improve properties and functionality using the electrical field. This check ends with signs of knowledge in upper grade Electrospinning production.
BACKGROUND:Prostate cancer (PCa) bone metastases cause significant morbidity and mortality in advanced disease. The tumor microenvironment (TME) of bone metastases drives disease progression and therapeutic resistance, yet comprehensive characterization of its cellular heterogeneity remains limited. This study aims to characterize cellular populations and molecular signatures of PCa bone metastases using single-cell RNA sequencing (scRNA-seq) data from the Gene Expression Omnibus (GEO) database. METHODS:scRNA-seq data from PCa bone metastasis samples were obtained from GEO. Quality control, normalization, dimensionality reduction, and cell type identification were performed using Seurat. Differential expression, pseudotime trajectory, pathway enrichment, gene regulatory network, and cell-cell communication analyses were conducted to investigate molecular mechanisms of bone metastasis progression. RESULTS:Single-cell analysis identified distinct cellular populations within the bone metastatic TME, including malignant epithelial cells, fibroblasts, endothelial cells, osteoblasts, osteoclasts, and immune cells. Clustering revealed heterogeneous transcriptional signatures, while pseudotime analysis uncovered developmental transitions between cell states. Key transcription factors, enriched pathways related to bone remodeling, angiogenesis, and immune regulation, and critical signaling interactions between cancer and stromal cells were identified. CONCLUSION:This study provides comprehensive insights into the cellular composition and molecular architecture of the PCa bone metastatic TME, revealing distinct cell populations, type-specific gene signatures, and cell-cell communication networks driving bone metastasis progression.
