Article
Computer Science, Theory & Methods
Zhongming Han, Xuelian Jin, Haozhen Xing, Weijie Yang, Haitao Xiong
Summary: Given the heterogeneity of real-world networks and the low efficiency of directly mining networks, learning low-dimensional embeddings of nodes in heterogeneous information networks (HINs) becomes crucial. In this paper, we propose a framework called HGSAGE for learning similarity-preserved embeddings of nodes in HINs. HGSAGE addresses the problems of omitting global information and considering only first-order neighbors by incorporating mechanisms to capture global information, sample and aggregate features from immediate and mediate neighbors, and combine embeddings from different meta-paths. Experimental results show that HGSAGE outperforms baseline methods on multiple tasks in real-world heterogeneous networks. Moreover, HGSAGE has important application values in this research field.
FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE
(2023)
Article
Computer Science, Information Systems
Xiaolu Zhang, Mingyuan Ma
Summary: Objectively evaluating representative papers in a specific scientific research field is important for the development of academia and scientific research institutions. The proposed GAEPIM framework uses a graph autoencoder based on heterogeneous networks to rank papers and find the most representative ones and their scientific institutions. The method outperforms several widely used baseline methods.
Article
Computer Science, Artificial Intelligence
Jie Zhang, Yishan Du, Pengfei Zhou, Jinru Ding, Shuai Xia, Qian Wang, Feiyang Chen, Mu Zhou, Xuemei Zhang, Weifeng Wang, Hongyan Wu, Lu Lu, Shaoting Zhang
Summary: The study proposes a graph-based method, DeepAAI, for predicting neutralization activity of antibodies and applies it to recommend probable antibodies for human immunodeficiency virus, severe acute respiratory syndrome coronavirus 2, influenza, and dengue. DeepAAI learns dynamic representations and relation graphs to optimize downstream tasks such as neutralization prediction and concentration estimation. The method demonstrates good performance and rich interpretability, suggesting potential broad-spectrum antibodies against new viral variants.
NATURE MACHINE INTELLIGENCE
(2022)
Article
Engineering, Biomedical
Jialin Wang, Shen Liang, Jiawei Zhang, Yingpei Wu, Lanying Zhang, Rui Gao, Dake He, C. -J. Richard Shi
Summary: Epilepsy is a common neurological disease, and deep learning models combined with graph neural network models have been used for single-channel EEG signal epilepsy detection. However, these methods lack interpretability of the classification results. To address this, researchers have attempted to combine graph representations of EEG signals with GNN models. Existing methods face challenges such as high time complexity in graph representations and the inability to integrate information from two domains. To overcome these challenges, a Weighted Neighbour Graph (WNG) representation is proposed, which is both time and space-efficient. A two-stream graph-based framework is also proposed to learn features from WNG in both time and frequency domains. Extensive experiments demonstrate the effectiveness and efficiency of the proposed methods.
IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Weishuai Che, Zhaowei Liu, Yingjie Wang, Jinglei Liu
Summary: The development of the Internet and big data has led to the importance of graphs as a data representation structure. However, as data size increases, graph embedding faces challenges in computational complexity and memory requirements. To address this, this paper proposes a multilevel embedding refinement framework (MERIT) based on large-scale graphs, using spectral distance-constrained graph coarsening algorithms and an improved graph convolutional neural network model. Experimental results show the effectiveness of MERIT, with an average AUROC score 8% higher than other baseline methods.
COMPLEX & INTELLIGENT SYSTEMS
(2023)
Article
Computer Science, Hardware & Architecture
Ru Huang, Lei Ma, Jianhua He, Xiaoli Chu
Summary: The passage introduces the definition of complex networks and their applications in real systems, proposes T-GAN as a new method for predicting temporal complex networks, and describes the framework and application instances of the method.
Article
Biochemical Research Methods
Chuanze Kang, Han Zhang, Zhuo Liu, Shenwei Huang, Yanbin Yin
Summary: This paper presents a novel GNN method LR-GNN based on link representation learning for accurately predicting molecular associations. Experimental results show that LR-GNN outperforms state-of-the-art methods and demonstrates robust ability to predict unknown associations. Visualizations also validate the effectiveness of the link representation used in LR-GNN.
BRIEFINGS IN BIOINFORMATICS
(2022)
Article
Computer Science, Artificial Intelligence
Babatounde Moctard Oloulade, Jianliang Gao, Jiamin Chen, Raeed Al-Sabri, Tengfei Lyu
Summary: This paper proposes a performance predictor-based graph neural architecture search (PGNAS) framework, which consists of three conceptually simpler and basic phases and can explore a search space with a cheaper computation cost. Experimental results show that PGNAS outperforms both handcrafted and Graph-NAS models on four benchmark datasets.
Article
Computer Science, Artificial Intelligence
Shengjie Min, Zhan Gao, Jing Peng, Liang Wang, Ke Qin, Bo Fang
Summary: Social Network Analysis has been widely used for intelligence gathering and criminal investigation by law enforcement agencies. Recent studies have focused on the application of Graph Neural Networks to solve social network problems, but there is a lack of research on time-evolving social networks, especially in the criminology field.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Yifan Zhou, learning Lei Yu
Summary: This paper proposes a new weighted prototype network for few-shot learning, which explores the contribution of each sample to its class prototype using graph neural networks. Experimental results show that the proposed model performs comparably to state-of-the-art approaches on three benchmark datasets.
PATTERN RECOGNITION LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Giorgos Bouritsas, Fabrizio Frasca, Stefanos Zafeiriou, Michael M. Bronstein
Summary: Although Graph Neural Networks (GNNs) have achieved remarkable results, recent studies have shown important shortcomings in their ability to capture the structure of the underlying graph. We propose Graph Substructure Networks (GSN), a topologically-aware message passing scheme based on substructure encoding, to address these limitations and obtain state-of-the-art results in various real-world settings.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Chemistry, Physical
Yang Zhong, Hongyu Yu, Xingao Gong, Hongjun Xiang
Summary: In this study, we propose a new framework called edge-based tensor prediction graph neural network that addresses the incompatibility of traditional invariant GNNs with directional properties. By expressing tensors as linear combinations of local spatial components projected on the edge directions of clusters with varying sizes, our framework is rotationally equivariant and satisfies the symmetry of local structures. We demonstrate the accuracy and universality of our framework by successfully predicting various tensor properties from first to third order. This work enables GNNs to step into the broad field of prediction of directional properties.
JOURNAL OF PHYSICAL CHEMISTRY LETTERS
(2023)
Article
Biology
Qing-Jing Sheng, Yuan Tan, Liyuan Zhang, Zhi-ping Wu, Beiying Wang, Xiao-Ying He
Summary: Long non-coding RNAs (lncRNAs) have crucial regulatory roles in cellular processes. Computational methods have emerged as valuable tools for predicting lncRNA functions and their associations with diseases. This study developed an improved computational method, RGLD, for predicting lncRNA-disease associations.
COMPUTERS IN BIOLOGY AND MEDICINE
(2023)
Article
Multidisciplinary Sciences
Luis M. C. Oliveira, Vinicius V. Santana, Alirio E. Rodrigues, Ana M. Ribeiro, Idelfonso B. R. Nogueira
Summary: This study developed a framework based on scientific machine learning strategies and utilized transfer learning to predict odor thresholds for chemical substances. Results showed that the transfer learning-based strategy displayed better predictive performance.
Article
Biochemical Research Methods
Wei Lan, Yi Dong, Qingfeng Chen, Ruiqing Zheng, Jin Liu, Yi Pan, Yi-Ping Phoebe Chen
Summary: This study proposes a computational method based on knowledge graph attention network to predict circRNA-disease associations. Experimental results show that this method outperforms other methods in predicting potential associations.
BRIEFINGS IN BIOINFORMATICS
(2022)
Article
Health Care Sciences & Services
Shahadat Uddin, Shangzhou Wang, Arif Khan, Haohui Lu
Summary: This study examines the progression of chronic diseases and their risk factors using a healthcare dataset sample of hospitalized patients. The results show that certain chronic diseases, such as cardiovascular diseases and diabetes, have a high prevalence in progressing to other chronic diseases, which is statistically significant. The progression frequencies increase with time and age, and the patients' sex also affects the disease progressions differently.
Article
Multidisciplinary Sciences
Shahadat Uddin, Stephen Ong, Petr Matous
Summary: Stakeholder engagement is a crucial factor affecting project outcomes, but there is a lack of empirical evidence on the differences in stakeholder engagement patterns between public, private, and public-private partnership (PPP) projects. This study uses social network research methods to capture and compare these engagement structures quantitatively. The findings reveal significant differences in network size, edge number, density, and betweenness centralization across the three types of projects. Additionally, the density varies significantly between 'within budget' and cost overrun projects for private and PPP projects. The study highlights the importance of network data and analytical techniques in managing relationships in complex project ecosystems.
Article
Computer Science, Artificial Intelligence
Taima Rahman Mim, Maliha Amatullah, Sadia Afreen, Mohammad Abu Yousuf, Shahadat Uddin, Salem A. Alyami, Khondokar Fida Hasan, Mohammad Ali Moni
Summary: Human Activity Recognition (HAR) is a valuable research field for clinical applications, where machine learning algorithms play a significant role. The proposed Gated Recurrent Unit-Inception (GRU-INC) model effectively utilizes both temporal and spatial information of time-series data, achieving high F1-scores on various publicly available datasets. The combination of GRU with Attention Mechanism and Inception module with Convolutional Block Attention Module (CBAM) contributes to the superior recognition rate and lower computational cost of the GRU-INC model. This framework has the potential to be applied in activity-associated clinical and rehabilitation applications.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Md Shofiqul Islam, Khondokar Fida Hasan, Sunjida Sultana, Shahadat Uddin, Pietro Lio', Julian M. W. Quinn, Mohammad Ali Moni
Summary: We propose a hybrid hierarchical attention-based bidirectional recurrent neural network with dilated CNN (HARDC) method for arrhythmia classification. This method fully exploits the dilated CNN and bidirectional recurrent neural network unit (BiGRU-BiLSTM) architecture to generate fusion features, improving the model's performance for prediction. By combining the fusion features with a dilated CNN and a hierarchical attention mechanism, the trained HARDC model showed significantly improved classification results and interpretability of feature extraction on the PhysioNet 2017 challenge dataset.
Article
Engineering, Industrial
Shahadat Uddin, Stephen Ong, Haohui Lu, Petr Matous
Summary: This study connects project management, network science, and machine learning by applying them to a real dataset. Relevant project data was collected through an online survey and categorized into three groups. Five machine learning approaches were used to model the relationships between project attributes, networks, and the Iron Triangle of project cost, time, and quality. The results confirm expected trends and provide an example for the applicability of integrated machine learning and network analytics to project performance modeling.
PRODUCTION PLANNING & CONTROL
(2023)
Article
Chemistry, Multidisciplinary
Nazim Choudhury, Shahadat Uddin
Summary: One of the characteristics of dynamic networks is the evolution of their actors and links. The link prediction mechanism in dynamic networks can capture the growth mechanisms of social networks. Researchers have utilized the temporal patterns of dynamic networks for dynamic link prediction. However, little attention has been given to the temporal variations of actor-level network structure and neighborhood information. This study attempts to build dynamic similarity metrics considering the temporal similarity and correlation between different actor-level evolutionary information of non-connected actor pairs. These metrics are used as dynamic features in the link prediction model and show improved performance compared to static similarity metrics.
APPLIED SCIENCES-BASEL
(2023)
Review
Health Care Sciences & Services
Palak Mahajan, Shahadat Uddin, Farshid Hajati, Mohammad Ali Moni
Summary: Machine learning models are utilized to create and improve disease prediction frameworks, and ensemble learning is a technique that combines multiple classifiers to enhance performance. In this study, the performance accuracies of different ensemble techniques (bagging, boosting, stacking, and voting) are assessed against five highly researched diseases. The findings reveal that stacking has the most accurate performance and can assist researchers in understanding current trends in disease prediction models that employ ensemble learning.
Review
Health Care Sciences & Services
Haohui Lu, Shahadat Uddin
Summary: This study presents a comprehensive review of graph machine learning methods and their applications in disease prediction using electronic health data. The commonly used approaches are shallow embedding and graph neural networks. While graph neural networks have shown outstanding results in disease prediction, they still face challenges in interpretability and dynamic graphs.
Article
Computer Science, Interdisciplinary Applications
Shahadat Uddin, Arif Khan, Haohui Lu
Summary: Research on COVID-19 has seen significant growth in recent years and has been a dominant topic in health-related publications. This study explores the impact of COVID-19 research on journal performance using the Impact Factor and six years of data. The results show that journals publishing COVID-19-related articles experienced a significant increase in their Impact Factor, with lower Impact Factor journals contributing the most to this growth. It suggests that journals prioritizing COVID-19 research may experience increased visibility and Impact Factor growth in the long term.
JOURNAL OF INFORMETRICS
(2023)
Article
Computer Science, Information Systems
Alireza Tavakolian, Alireza Rezaee, Farshid Hajati, Shahadat Uddin
Summary: The study presents a hybrid deep model, GAOCNN, for predicting hospital readmission and length of stay. The model utilizes one-dimensional convolutional layers and optimizes the layer parameters through a genetic algorithm. Experimental results demonstrate the high accuracy of the model in predicting readmission and length of stay for patients with various conditions. This research provides a platform for managing healthcare resources.
Article
Health Care Sciences & Services
Md. Martuza Ahamad, Sakifa Aktar, Md. Jamal Uddin, Md. Rashed-Al-Mahfuz, A. K. M. Azad, Shahadat Uddin, Salem A. Alyami, Iqbal H. Sarker, Asaduzzaman Khan, Pietro Lio, Julian M. W. Quinn, Mohammad Ali Moni
Summary: Good vaccine safety and reliability are crucial for countering infectious diseases effectively. This study aims to reduce adverse reactions to COVID-19 vaccines by identifying common factors through patient data analysis and classification. Patient medical histories and postvaccination effects were examined, and statistical and machine learning approaches were used. The analysis revealed that prior illnesses, hospital admissions, and SARS-CoV-2 reinfection were significantly associated with poor patient reactions.
Proceedings Paper
Computer Science, Theory & Methods
Fangyu Zhou, Shahadat Uddin
Summary: This paper introduces a network-based approach using graph neural networks to support the early detection of adverse drug events. The method models each patient as a subgraph and achieves high accuracy and recall in identifying cohorts associated with adverse drug events.
PROCEEDINGS OF 2023 AUSTRALIAN COMPUTER SCIENCE WEEK, ACSW 2023
(2023)
Proceedings Paper
Computer Science, Theory & Methods
Fangyu Zhou, Shahadat Uddin
Summary: In recent years, there has been an exponential growth in drug-related data and adverse drug reactions (ADRs), leading to a comparatively high hospitalization rate worldwide. To minimize risks, extensive research has been conducted to predict ADRs. Due to the high cost and time-consuming nature of lab experiments, researchers are exploring the use of data mining and machine learning techniques in this field. This paper constructs a weighted drug-drug network by integrating various data sources, revealing underlying relationships between drugs based on common ADRs. Network features are extracted from this network, such as weighted degree centrality and weighted PageRanks, which are concatenated with original drug features to train and test seven classical machine learning algorithms. Experiment results show that adding these network measures benefits all tested machine learning methods, with logistic regression achieving the highest mean AUROC score (0.821) across all ADRs. Weighted degree centrality and weighted PageRanks are identified as the most important network features in the logistic regression classifier. This evidence strongly supports the fundamental role of the network approach in future ADR prediction, where network edge weights play a crucial role in the logistic regression model.
PROCEEDINGS OF 2023 AUSTRALIAN COMPUTER SCIENCE WEEK, ACSW 2023
(2023)