Graph-level anomaly detection is a critical yet un-derexplored task, especially in the neuroimaging domain where early identification of abnormal brain patterns is vital. In this paper, we propose KDGAE, a generalizable graph autoencoder framework based on knowledge distillation. It employs a teacher-student architecture for graph reconstruction and latent representation distillation. Anomaly scores are then derived from both reconstruction errors and embedding discrepancies. To improve interpretability, post-hoc explanation tools such as feature masking and GNNExplainer are integrated. KDGAE is evaluated on the UB-GOLD benchmark and the Autism Brain Imaging Data Exchange (ABIDE) dataset for autism detection, achieving competitive performance against state-of-the-art methods under limited supervision.

The identification of cancer driver genes is important for better understanding the hallmarks of cancer and developing precision therapies. Though the integration of multiomics and protein-protein interaction (PPI) data into graph convolutional networks (GCN) has been emerging as a promising strategy, these GCN-based methods rely heavily on the reliability and completeness of the PPI network data, thereby hampering the identification of pan-cancer genes and key oncogenic interactions. Furthermore, few GCN-based models today enables the detection of function modules and driver genes in a simultaneous manner. To this end, we introduce a novel GCN-based model, "Gated graph convolutional network with ATtention mechanism for identifying cancer Driver Genes and function modules (DGGAT)". This model integrates the gating mechanisms with Gumbel-Softmax re-parameterization and attention mechanisms, aiming to control which interactions participate in the flow of information in PPI and weight the interactions. We apply DGGAT to identify novel cancer genes and function modules through five PPI networks. Comparison results of our model with baseline models indicated that our model obtain the highest accuracy in cancer genes identification. Further case study demonstrated that the model possesses the capability to unveil the molecular mechanism with its inherent function modules.

Graph sampling frequently compresses a large graph into a limited screen space. This paper proposes a hierarchical structure model that partitions scale-free graphs into three blocks: the core, which captures the underlying community structure, the vertical graph, which represents minority structures that are important in visual analysis, and the periphery, which describes the connection structure between low-degree nodes. A new algorithm named hierarchical structure sampling (HSS) was then designed to preserve the characteristics of the three blocks, including complete replication of the connection relationship between high-degree nodes in the core, joint node/degree distribution between high- and low-degree nodes in the vertical graph, and proportional replication of the connection relationship between low-degree nodes in the periphery. Finally, the importance of some global statistical properties in visualization was analyzed. Both the global statistical properties and local visual features were used to evaluate the proposed algorithm, which verify that the algorithm can be applied to sample scale-free graphs with hundreds to one million nodes from a visualization perspective.

Linked Open Data (LOD) is an ongoing effort in the Semantic Web community to build a massive public knowledge graph. The goal is to extend the Web by publishing various open datasets as RDF on the Web and then linking data items to other useful information from different data sources. With linked data, starting from a certain point in the graph, a person or machine can explore the graph to find other related data. In this paper, we develop a novel pipeline for graph-based biological data integration. By using our pipeline, users can easily glue heterogeneous biological ontologies, annotate sources with multiple join tables effectively, obtain a high-quality biological knowledge graph automatically, and enrich the knowledge graph with public biological ontologies finally. We implement a platform that realizes the proposed approach and conduct two case studies to evaluate the effectiveness and efficiency of our approach.

Bridging heterogeneous mutation data fills in the gap between various data categories and propels discovery of disease-related genes. It is known that genome-wide association study (GWAS) infers significant mutation associations that link genotype and phenotype. However, due to the differences of size and quality between GWAS studies, not all de facto vital variations are able to pass the multiple testing. In the meantime, mutation events widely reported in literature unveil typical functional biological process, including mutation types like gain of function and loss of function. To bring together the heterogeneous mutation data, we propose a 'Gene-Disease Association prediction by Mutation Data Bridging (GDAMDB)' pipeline with a statistic generative model. The model learns the distribution parameters of mutation associations and mutation types and recovers false-negative GWAS mutations that fail to pass significant test but represent supportive evidences of functional biological process in literature. Eventually, we applied GDAMDB in Alzheimer's disease (AD) and predicted 79 AD-associated genes. Besides, 12 of them from the original GWAS, 60 of them are supported to be AD-related by other GWAS or literature report, and rest of them are newly predicted genes. Our model is capable of enhancing the GWAS-based gene association discovery by well combining text mining results. The positive result indicates that bridging the heterogeneous mutation data is contributory for the novel disease-related gene discovery.

MOTIVATION: Drug-drug interactions (DDIs) are one of the major concerns in pharmaceutical research. Many machine learning based methods have been proposed for the DDI prediction, but most of them predict whether two drugs interact or not. The studies revealed that DDIs could cause different subsequent events, and predicting DDI-associated events is more useful for investigating the mechanism hidden behind the combined drug usage or adverse reactions. RESULTS: In this article, we collect DDIs from DrugBank database, and extract 65 categories of DDI events by dependency analysis and events trimming. We propose a multimodal deep learning framework named DDIMDL that combines diverse drug features with deep learning to build a model for predicting DDI-associated events. DDIMDL first constructs deep neural network (DNN)-based sub-models, respectively, using four types of drug features: chemical substructures, targets, enzymes and pathways, and then adopts a joint DNN framework to combine the sub-models to learn cross-modality representations of drug-drug pairs and predict DDI events. In computational experiments, DDIMDL produces high-accuracy performances and has high efficiency. Moreover, DDIMDL outperforms state-of-the-art DDI event prediction methods and baseline methods. Among all the features of drugs, the chemical substructures seem to be the most informative. With the combination of substructures, targets and enzymes, DDIMDL achieves an accuracy of 0.8852 and an area under the precision-recall curve of 0.9208. AVAILABILITY AND IMPLEMENTATION: The source code and data are available at https://github.com/YifanDengWHU/DDIMDL. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

After mutation, a gene either gains or loses a function, which is considered to be a critical piece of information for understanding a related pathology and thus for drug discovery. By classifying association of genes and phenotypes using the information of gain-of-function or loss-of-function, the exploration for drug repurposing can be guided in a much more efficient way. We present the active gene annotation corpus (AGAC), which contains 500 manually annotated abstracts collected from PubMed. Five bio-concept labels and three regulatory concept labels were designed for concept level annotation. Furthermore, two relation types, “Theme” and “Cause”, were designed to interlink regulatory concepts with their thematic and causal elements. We evaluated AGAC from three aspects, the results of which indicate the high quality of the annotation. Eventually, a PubMed-wide case study was performed to show that by using AGAC the process of anti-epilepsy drug discovery can be enhanced. After text retrieval, filtering, text classification and matching with DrugBank entries, 281 gene-drug pairs and 112 drugs were obtained and 30 out of 112 were recorded in databases. Among 10 newly predicted multi-target drugs which were not recorded in databases, 6 of them were found to be related to epilepsy with literature support, i.e., Oxazepam, Temazepam, Halazepam, Prazepam, Zolpidem and Thiamylal. The result of the case study support the potential of AGAC for enhancing knowledge discovery for drug repurposing.

The non‐equilibrium of network traffic data brings about the non‐equilibrium of classification. Feature extraction is an effective method to reduce data dimensions, while it can intensify the influence of non‐equilibrium further. A secondary feature extraction algorithm of multidimensional assessment is proposed in this study. The features of network traffic are evaluated in different dimensions to provide the basis for feature extraction. Furthermore, a model dealing with imbalanced data is proposed based on secondary feature extraction and sampling. The model combines the benefits of dimension reduction and redistribution. The experiment results show that the proposed model can not only increase classification accuracy and decrease non‐equilibrium, but also enhance the performance of different classification algorithms.