Polypharmacy has become essential in managing complex and chronic conditions, yet it introduces significant clinical risks due to potential drug-drug interactions (DDIs). Existing computational models struggle to provide accurate and interpretable predictions, largely due to fragmented integration of molecular structures and biomedical knowledge. These limitations arise from static graph designs, inadequate substructure modeling, and insufficient incorporation of semantic context, highlighting the need for a unified, explainable solution. In this study, we propose MKGFlow-DDI, a multi-view knowledge-guided framework that jointly leverages drug-drug interaction networks and biomedical knowledge graphs to dynamically construct drug-flow subgraphs. The model incorporates a dual-channel encoder designed to capture atom-level information and substructure-level features, integrating them with the global semantic embeddings of the composite network to derive novel feature representations. These representations are fused to initialize node features within each subgraph, which are iteratively optimized through similarity-based edge refinement to reduce noise and enhance biological relevance. To further improve generalization and stability, a contrastive learning module is introduced to align representations of perturbed subgraphs by maximizing consistency across positive and negative sample pairs. Experimental results on DrugBank and TWOSIDES demonstrate that MKGFlow-DDI outperforms state-of-the-art baselines, especially in scenarios involving previously unseen drugs. Additionally, the model produces interpretable semantic pathways that align with known pharmacological mechanisms, enabling clinically meaningful insights. Overall, MKGFlow-DDI establishes a robust and biologically grounded approach to DDI prediction, offering a promising direction for computational pharmacovigilance and personalized therapy optimization.
The prediction of a biological sequence's function is an exploding field that could prove to be an invaluable tool in deciphering molecular mechanisms of diseases and genetic variation. However, existing techniques suffer from a very low number of experimentally verified labels as well as the complexity of modeling sequence context. Previous work only focuses on semi-supervised learning (SSL) or pre-trained language models (PLMs) individually and does not consider their complementarity: whereas PLMs capture syntactic conservatism and long-range dependency from huge unlabeled sequence data as features, SSL updates label-level certainty via confidence-weighted pseudo-label selection. To realize integrating these benefits, we make a creative use of PLMs as strong feature extractors to make the sequence semantics extracted, and meanwhile leverage the SSL to constrain the decision boundary by selecting reliable pseudo-labels. We show that our framework achieves competitive performance compared to full supervision while using far fewer labeled samples, on two biological prediction tasks: DNA-binding protein (DBP) and non-coding RNA (ncRNA) detection. In general, our approach, developed here, not only offers an efficient solution for discovering new DBP and ncRNA in low-resource settings where experimental verification is infeasible, but also lays a solid methodological foundation for biological sequence classification in computational biology.Key points Unlabeled data significantly improves model accuracy in semi-supervised learning. The language model-based self-training method outperforms traditional semi-supervised approaches like TSVM. It achieves competitive performance against state-of-the-art fully supervised methods, requiring only 1% labeled data for effective classification and accurately identifying novel biomolecules.
Emerging evidence implicates peripheral metabolic disturbances in the pathogenesis of cerebrovascular diseases. However, the specific molecular conduits linking angina pectoris (AP) to ischemic stroke (IS) remain elusive. To address this, we evaluated the causal effect of AP on IS subtypes utilizing Bayesian-weighted Mendelian randomization (BWMR) based on large-scale GWAS meta-data (n = 628 000). Cross-tissue transcriptomic analyses, coupled with the Boruta feature selection algorithm, were employed to screen for candidate mediators, with subsequent validation of genetic causality and therapeutic viability via standard Mendelian randomization (MR), colocalization analysis, and molecular docking. Furthermore, their involvement in cellular immunity was characterized using CIBERSORT and single-cell ribonucleic acid-seq (scRNA-seq). BWMR established a robust causal relationship from stable AP to IS (OR = 1.13, P = .004). Multi-omics integration identified 18 hub macromolecules, prioritizing HMGCR, TLR4, and MMP9 as the primary candidate mediators enriched in lipid metabolism and inflammatory pathways. MR corroborated the genetic relevance of these loci to IS risk, while molecular docking confirmed their structural druggability. A multilayer perceptron classifier (MLPC) utilizing these targets demonstrated high predictive accuracy for IS (AUC = 0.82-0.90). Crucially, scRNA-seq revealed that TLR4 and MMP9 expression in peripheral immune cells (CD8+ T cells and neutrophils) may drive a metabolic-inflammatory cascade, functioning as vital conduits for peripheral-to-brain signal transduction. In conclusion, this multi-omics framework identifies HMGCR, TLR4, and MMP9 as genetically supported mediators bridging AP and IS, elucidating peripheral-cerebral crosstalk and highlighting actionable targets for clinical risk stratification and pharmacological repurposing.
Cancer-associated fibroblasts (CAFs) activated by intercellular communication contribute to the progression of esophageal squamous cell carcinoma (ESCC). Migrasomes represent a novel mode of intercellular communication. However, the characteristics of ESCC-derived migrasomes in tumor hypoxic microenvironments and their effects on CAFs remain unclear. Migrasomes were isolated from ESCC cells under normoxia/hypoxia, with TSPAN4-GFP labeling, transmission electron microscopy, nanoparticle tracking analysis, and western blot for validation. Whole-transcriptome sequencing analyzed hypoxic migrasome RNA profiles, and ceRNA networks were predicted via RNAhybrid and Miranda. The effect of migrasomes on CAFs was assessed using fluorescence tracing, reverse transcription‑quantitative polymerase chain reaction, Transwell migration, enzyme‑linked immunosorbent assay, and western blot. ESCC cells produced migrasomes. Although hypoxia did not alter their quantity or structure, it significantly altered their RNA cargo, changing the composition of mRNA, lncRNA, and circRNA. Differentially expressed mRNAs were enriched in "Response to hypoxia" and "HIF-1 signaling pathway." DElncRNAs were enriched in "Golgi to plasma membrane protein transport" and "Cell adhesion molecules pathway," while DEcircRNAs were enriched in "ubiquitin binding" and "chromatin remodeling." Predicted ceRNA networks were constructed using RNAhybrid and Miranda, involving 659 miRNAs, 24 lncRNAs, and 132 mRNAs. Fibroblasts internalized migrasomes and acquired a CAF-like phenotype, showing enhanced migration, elevated secretion of IL-1β/TGF-β, and increased CAF marker expression (α-SMA, COL1A1, COL3A1, FAP, PDGFRβ), with effects most pronounced under hypoxic migrasome treatment. This study characterized hypoxic migrasome whole transcriptome landscapes and suggested that hypoxic migrasomes may promote CAF-like changes in vitro, uncovering a novel ESCC-tumor microenvironment interaction mechanism and offering new perspectives for ESCC research.
RNA viruses cause substantial global disease burden and depend on host RNA-binding proteins and translation machinery. However, it remains unclear which host factors are robustly engaged across independent Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) RNA interactome studies and to what extent these factors are shared with other RNA viruses. Here, we perform an integrative meta-analysis of eight published SARS-CoV-2 RNA-protein interactomes and compare them with corresponding Influenza A virus, Zika virus, and Dengue virus datasets to define conserved host networks and prioritize candidate host-directed antiviral targets. By integrating multiple datasets and applying ClusterProfiler together with curated pathway resources (KEGG, Reactome, WikiPathways, and Gene Ontology), we systematically characterize the functional landscape of SARS-CoV-2 RNA-protein interactions. The consensus SARS-CoV-2 interactome is enriched for mRNA processing, translation, RNA surveillance and innate immune functions. Cross-viral comparison identifies 275 host proteins shared across all four RNA viruses, forming interconnected modules that include key translation factors (EEF1A1, EIF4A1, EIF3H) and RNA-binding proteins (Nucleolin, ILF3). Drug-target annotation prioritizes 21 proteins with 35 approved or investigational modulators for host-directed antiviral repurposing. Together, these findings generate a consensus map of conserved host dependencies and highlight prioritized targets for future mechanistic and translational studies.Research Highlights Integrated SARS-CoV-2 datasets and compared with, Influenza A virus, Zika virus, Dengue virus. Identified 275 host proteins shared across these four pathogens. Conserved proteins were enriched in translation, RNA processing, and innate immune pathways. Prioritized 21 host targets and 35 drugs for antiviral repurposing.
