Protein complexes play essential roles in cellular functions, and accurate identification of these complexes is critical for understanding biological processes and disease mechanisms. Existing methods frequently compromise the global topology of protein-protein interaction (PPI) networks when incorporating biological resources. Moreover, they fail to adequately address the intrinsic sparsity of PPI data and the widespread occurrence of false positives and false negatives. These approaches also struggle to capture the diverse neighborhood dependencies necessary to represent distinct functional roles of proteins within complexes. To address these limitations, we propose a knowledge-enhanced graph contrastive learning (KEGCL) framework for protein complex identification. KEGCL constructs a knowledge-enhanced PPI network by integrating external biological priors. A perturbation strategy guided by spatiotemporal constraints is then applied to selectively reintroduce functionally relevant interactions, thereby enhancing semantic diversity in the generated graph views. Based on this, graph convolutional encoders with randomized propagation depths are used to capture protein interaction patterns at multiple structural levels, enhancing the model's ability to represent both densely connected cores and loosely associated attachments within protein complexes. Extensive experiments on multiple real-world PPI datasets show that KEGCL achieves competitive performance compared with state-of-the-art methods, and enrichment analyses confirm the biological relevance of the identified complexes.

As an effective and low-risk approach to identify new therapeutic pathways for existing drugs, drug repositioning has been extensively utilized to expedit drug discovery processes. However, current knowledge graph (KG)-based methodologies encounter several hurdles in this context. Firstly, most graph neural network (GNN)-based approaches fail to adequately capture the intricate relationships between drug-drug, drug-disease, or disease-disease. Secondly, the subtle synergistic mechanisms between drugs and diseases remain underexplored. Lastly, the training of knowledge graph embedding (KGE) methods is susceptible to noise, leading to unstable model optimization. To address these challenges, we intruduce KRANE, a knowledge-driven and relation-aware synergistic learning method for drug repositioning. KRANE addresses these issues through three innovative modules. Firstly, we design a relation-aware feature extractor (RAFE), which utilizes the contextual triples attention scores in KG to effectively integrate drug-related knowledge and enhance the representation of complex relational features. Secondly, we adopt a synergistic feature reconstruction module as a decoder to extract synergistic heterogeneous feature interactions between drugs and diseases from entity and relation representations. Finally, we propose a knowledge-regulated loss function to mitigate the impact of noise on model training. Experiments conducted on three publicly available datasets demonstrate that KRANE significantly outperforms existing methods.

Identifying biologically significant protein complexes from protein-protein interaction (PPI) networks and understanding their roles are essential for elucidating protein functions, life processes, and disease mechanisms. Current methods typically rely on static PPI networks and model PPI data as pairwise relationships, which presents several limitations. Firstly, static PPI networks do not adequately represent the scopes and temporal dynamics of protein interactions. Secondly, a large amount of available biological resources have not been fully integrated. Moreover, PPIs in biological systems are not merely one-to-one relationships but involve higher order non-pairwise interactions. To alleviate these issues, we propose HGST, a multi-source biological knowledge-guided hypergraph spatiotemporal subnetwork (subnet) embedding method for identifying biologically significant protein complexes from PPI networks. HGST initially constructs spatiotemporal PPI subnets using the scopes and temporal dynamics of proteins derived from multi-source biological knowledge, treating them as dynamic networks through fine-grained spatiotemporal partitioning. The spatiotemporal subnets are then transformed into hypergraphs, which model higher order non-pairwise relationships via hypergraph embedding. Simultaneously, fine-grained amino acid sequence features and coarse-grained gene ontology attributes are introduced for multi-dimensional feature fusion. Finally, protein complexes are identified from the reweighted subnets based on fused feature representations using the core-attachment strategy. Evaluations on four real PPI datasets demonstrate that HGST achieves competitive performance. Furthermore, a series of biological analyses confirm the high biological significance of the complexes identified by HGST. The source code is available at https://github.com/qifen37/HGST.

Drug repositioning, which identifies new therapeutic potential of approved drugs, is instrumental in accelerating drug discovery. Recently, to alleviate the effect of data sparsity on predicting possible drug-disease associations (DDAs), graph contrastive learning (GCL) has emerged as a promising paradigm for learning discriminative representations of drugs and diseases through distilling informative self-supervised signals. However, existing GCL-based methods devised for DDA prediction still encounter two limitations. Firstly, the crucial heterogeneous property, which allows for capturing nuanced interaction semantics between biological entities, is overlooked. The second is how to perform contrastive view augmentation without relying on stochastic perturbation. In this study, we propose a novel multi-view contrastive learning approach for DDA prediction, namely MICLE. To handle the first issue, protein-related bipartite graphs are integrated with the original DDA network in advance, thereby composing a heterogeneous biological network (HBN). Besides, heterogeneous graph neural network is applied to mine the rich connectivity patterns implicit in the above HBN. For the second limitation, we design the complementary inter-view and intra-view contrastive learning tasks. Specifically, the former ensures that the mutual information between paired nodes across views is maximized, the latter enhances the agreement between each node and its first-order neighbors on similarity networks. Extensive experiments conducted on three benchmarks under 10-fold cross-validation demonstrate the model effectiveness.

The identification of protein complexes from protein interaction networks is crucial in the understanding of protein function, cellular processes and disease mechanisms. Existing methods commonly rely on the assumption that protein interaction networks are highly reliable, yet in reality, there is considerable noise in the data. In addition, these methods fail to account for the regulatory roles of biomolecules during the formation of protein complexes, which is crucial for understanding the generation of protein interactions. To this end, we propose a SpatioTemporal constrained RNA-protein heterogeneous network for Protein Complex Identification (STRPCI). STRPCI first constructs a multiplex heterogeneous protein information network to capture deep semantic information by extracting spatiotemporal interaction patterns. Then, it utilizes a dual-view aggregator to aggregate heterogeneous neighbor information from different layers. Finally, through contrastive learning, STRPCI collaboratively optimizes the protein embedding representations under different spatiotemporal interaction patterns. Based on the protein embedding similarity, STRPCI reweights the protein interaction network and identifies protein complexes with core-attachment strategy. By considering the spatiotemporal constraints and biomolecular regulatory factors of protein interactions, STRPCI measures the tightness of interactions, thus mitigating the impact of noisy data on complex identification. Evaluation results on four real PPI networks demonstrate the effectiveness and strong biological significance of STRPCI. The source code implementation of STRPCI is available from https://github.com/LI-jasm/STRPCI.

Automated International Classification of Diseases (ICD) coding tasks are designed to assign diagnosis and procedure codes to patients’ electronic medical records (EMRs). Recent works have applied deep neural network models and related techniques for code assignment to clinical notes. However, most existing methods have overlooked the advantageous complementary information presented in the tabular data from EMRs and the Wikipedia knowledge. Therefore, we propose a Multimodal Knowledge Fusion Network (MKFN) to effectively integrate clinical notes, tabular data, and Wikipedia knowledge, enhancing the model’s predictive capabilities. We incorporate structured tabular data and clinical notes into an initial multimodal representation using label attention and self-attention mechanisms. We propose a knowledge fusion network to leverage tabular data and Wikipedia knowledge for accurate predictions when code descriptions are absent in clinical notes. Experiments on the MIMIC dataset show that our proposed model achieves competitive results among existing ICD coding methods.