patient-specific network encoding

Accurately predicting prognosis for intracranial aneurysm patients remains challenging due to the dynamic, heterogeneous nature of clinical trajectories. Method: We propose modeling clinical care pathways as a temporal knowledge graph (TKG) and learning personalized, time-aware patient representations via graph neural networks. Specifically, we construct a multi-granularity TKG integrating clinical entities, relations, and timestamps; incorporate multi-strategy temporal encoding with graph convolutional network (GCN) embedding; and systematically evaluate TKG-based prediction against conventional tabular models. Contribution/Results: Our approach achieves statistically significant AUC improvement over standard baselines, providing the first systematic validation of TKGs for clinical prognosis prediction. We identify that literal value representation of entity attributes and TKG schema design critically influence performance—establishing an empirically optimal graph structure. Quantitative analysis further reveals diminishing returns from increasingly complex temporal encodings. These findings underscore the unique value and practical potential of TKGs in biomedical predictive modeling.

Existing methods for predicting critical illness severity in ICU patients neglect inter-patient relational structures. Method: We propose a dynamic self-constructing graph neural network (GNN) that dynamically builds patient similarity graphs by jointly leveraging feature and structural similarity. A hybrid GNN architecture—integrating GCN, GraphSAGE, and GAT—jointly models local neighborhood patterns and global topological structures; an attention mechanism enables adaptive weighting of clinically relevant features and enhances prediction interpretability. Results: Evaluated on the MIMIC-III dataset (6,000 ICU admissions), our model achieves an AUC-ROC of 0.94, significantly outperforming baseline and single-GNN models. It also improves recall and provides clinically interpretable predictions, establishing a new paradigm for early critical illness warning that balances accuracy and clinical trustworthiness.

Medical AI systems frequently fail during cross-institutional deployment due to distributional shift, discriminatory bias, and lack of transparency—rooted in reliance on spurious statistical associations rather than invariant causal mechanisms. To address this, we propose the Causal Graph Neural Network (CGNN) framework, which integrates biomedical graph structures with structural causal models to disentangle and learn invariant disease–treatment causal mechanisms. CGNN innovatively separates causal-inspired modeling from causal validation, enabling counterfactual reasoning and intervention prediction. It further incorporates large language models to assist in causal hypothesis generation and mechanism verification. We validate CGNN across four clinical scenarios: brain-network-based psychiatric diagnosis, multi-omics cancer subtyping, physiological monitoring explanation, and debiased drug recommendation. Results demonstrate substantial improvements in out-of-distribution generalization, algorithmic fairness, and model interpretability. This work establishes a methodological foundation for trustworthy, patient-specific causal digital twins.

Biological pathway graphs exhibit high topological complexity and suffer from severe distortion when embedded in Euclidean space. To address this, we propose MC-GCN—the first non-Euclidean graph neural network that learns pathway node embeddings on a product manifold endowed with mixed curvature. Methodologically, MC-GCN integrates multi-curvature geometric modeling with product manifold optimization, designs a curvature-aware graph convolution tailored for highly distorted structures, and employs a supervised edge-prediction framework. Its key contribution lies in pioneering the incorporation of mixed-curvature geometry and product manifold representation into pathway embedding learning, effectively mitigating embedding distortion caused by global curvature inconsistency. Experiments demonstrate that MC-GCN significantly reduces embedding distortion and achieves substantial improvements in accuracy on in-distribution protein–protein interaction prediction. The source code and pathway analysis toolkit are publicly available.

This study investigates how topological properties of biomedical knowledge graphs (KGs) influence link prediction performance. We systematically analyze structural characteristics—including sparsity, degree distribution, clustering coefficient, and relation symmetry—across benchmark datasets (e.g., DrugBank, Hetionet), and evaluate their impact on KG completion using representative embedding models (TransE, RotatE, ComplEx) via controlled ablation studies. Our key contribution is the first empirical quantification of associations between graph-structural metrics and link prediction accuracy (measured by MRR and mean rank), achieving an R² of 0.73. We identify local clustering coefficient and relation symmetry as the most predictive topological factors. To ensure reproducibility and facilitate structural attribution analysis, we publicly release all prediction results and a dedicated analytical toolkit. This work establishes interpretable, topology-aware design principles for biomedical KG modeling, bridging structural graph theory with practical knowledge representation tasks.

研究探讨了在生物医学知识图谱中使用双曲图表示学习进行孟德尔疾病鉴别诊断，实验表明该方法能在较低维度上有效利用层次结构并支持异质患者级图的诊断推理。

High-resolution 3D modeling of vascular structures faces dual challenges of geometric complexity and computational efficiency. This work proposes the first graph tokenization framework tailored for tubular anatomical structures, leveraging centerline points and pseudo-radii to construct a neural implicit representation that encodes both geometric and topological information into compact, generalizable latent tokens. The approach enables efficient solutions to inverse problems such as reconstruction, generation, and link prediction. Extensive experiments on pulmonary airways, pulmonary vasculature, and cerebral vasculature demonstrate the method’s strong cross-anatomical generalization capability and anatomically plausible generation performance.

为解决患者个性化医疗概念表示学习问题，提出REFINE方法，通过预算限制下的文本属性图和强化学习策略选择KG上下文，并用LLM进行语义优化。

This work addresses the collapse of feature representations in conventional medical deep learning models caused by task-specific training, which undermines transferability, stability, and interpretability. To overcome this limitation, the authors propose a dense feature learning framework that, for the first time, treats the geometric structure of representations as a first-class optimization objective in medical AI. By directly optimizing linear algebraic properties—such as spectral balance, subspace consistency, and feature orthogonality—on the embedding matrix, the method explicitly preserves the intrinsic linear structure of clinical data without requiring labels or reconstruction. Evaluated on longitudinal electronic health records, clinical text, and multimodal patient representations, the approach significantly outperforms both supervised and self-supervised baselines, demonstrating consistent improvements in downstream linear performance, robustness, and subspace alignment.

This study addresses the significant variability in breast cancer patients’ response to neoadjuvant chemotherapy (NACT) and the urgent need for accurate prediction of pathological complete response (pCR). To this end, the authors propose a novel approach that, for the first time, employs a 3D spatiotemporal graph neural network to model longitudinal dynamic contrast-enhanced MRI (DCE-MRI) data. Their method explicitly captures temporal interactions across multiple imaging timepoints and incorporates three complementary self-supervised learning objectives to enable personalized treatment response prediction. Evaluated on the ISPY-2 dataset comprising 585 patients, the proposed framework substantially outperforms existing visual and self-supervised baselines, establishing a new state-of-the-art benchmark for pCR prediction. The authors further release their code and data processing library to support reproducible research in this domain.