Transformers, with their self-attention mechanisms and positional encoding, excel at modeling long-range dependencies. Such attribute has demonstrated significant potential in capturing complex disease patterns by integrating multimodal information, for example clinical notes and radiographs. However, their reliance on pre-trained deep neural networks to extract modality-specific features from large datasets makes them vulnerable to backdoor attacks, posing critical challenges for their deployment in healthcare applications. To address these vulnerabilities, we propose a robust multimodal Transformer-based framework, RMTrans, which mitigates the impact of malicious imaging data containing backdoor triggers while enhancing the model's robustness. In the imaging data pre-processing stage, we introduce an efficient patch-based processing method that shifts the model's focus toward learning global features rather than over-fitting to localized (patch-level) patterns, thereby ensuring a more secure and reliable training process. Following this, we fuse multimodal representations and train a Vision Transformer (ViT) for disease prediction. Extensive experiments conducted on real-world datasets, including MIMIC-IV and CXR, validate the effectiveness of RMTrans. The proposed framework outperforms state-of-the-art baselines, demonstrating its potential as a secure and reliable solution for multimodal disease prediction.

With the rapid advancements in large language model technology and the emergence of bioinformatics-specific language models (BioLMs), there is a growing need for a comprehensive analysis of the current landscape, computational characteristics, and diverse applications. This survey aims to address this need by providing a thorough review of BioLMs, focusing on their evolution, classification, and distinguishing features, alongside a detailed examination of training methodologies, datasets, and evaluation frameworks. We explore the wide-ranging applications of BioLMs in critical areas such as disease diagnosis, drug discovery, and vaccine development, highlighting their impact and transformative potential in bioinformatics. We identify key challenges and limitations inherent in BioLMs, including data privacy and security concerns, interpretability issues, biases in training data and model outputs, and domain adaptation complexities. Finally, we highlight emerging trends and future directions, offering valuable insights to guide researchers and clinicians toward advancing BioLMs for increasingly sophisticated biological and clinical applications.

Few-shot Knowledge Graph Completion (FKGC) has emerged as a significant area of interest for addressing the long-tail problem in knowledge graphs. Traditional approaches often focus on the sparse few-shot neighborhood to derive semantic representation, overlooking other critical information forms such as relation paths. In this paper, we introduce an innovative method, called PARE, which fully leverages relation paths to enhance the few-shot representation by simultaneously incorporating both neighborhood and relation path information. Inspired by the principles of information transmission, PARE directly models relation paths between entities and parameterizes the information interference within different relation paths. Through parameter learning, PARE effectively captures information propagation along relation paths while mitigating the influence of relation dependency. To preserve neighborhood information, we employ a two-step neighborhood aggregator to resolve few-shot neighbors’ ambiguity and develop a reconstruction module. By integrating the representations of relation paths and contextual neighborhoods, we achieve a comprehensive fewshot representation for two given entities. We utilize a matching processor for knowledge triplet evaluation. Extensive experiments demonstrate that our PARE model outperforms state-of-the-art baselines on widely-used benchmark datasets.

Multiscale brain networks are crucial for diagnosing brain disorders by revealing the hierarchical organization of brain function and connectivity. However, previous methods that explored multi-atlas approaches to model these networks often failed to represent this organization across multiple spatial and temporal scales, leading to limited representations and potentially inaccurate diagnoses. To address this issue, we propose the Multiscale Brain Graph Transformer (BrainMGT), which captures the hierarchical organization of brain connectivity at various spatial and temporal scales to improve the diagnosis of brain disorders. BrainMGT constructs multiscale brain networks that model spatial hierarchies, i.e., microscale, mesoscale, and macroscale, while preserving their modular structure. It also incorporates multiple temporal scales, i.e., fast, intermediate, and slow, for estimating connectivity, rather than relying on a single temporal scale from blood-oxygen-level-dependent (BOLD) signals. Using self-attention and cross-attention mechanisms, BrainMGT extracts and integrates features both within and between these scales, generating fine-coarse feature representations that improve diagnostic precision. We validated BrainMGT on three real-world functional magnetic resonance imaging (fMRI) datasets, and the results show that BrainMGT outperformed existing methods in diagnosing neurological disorders.

Knowledge graphs (KGs) serve as structured representations of knowledge, comprising entities and relations. KGs are inherently incomplete, sparse, and have a strong need for completion. Although many knowledge graph embedding models have been designed for knowledge graph completion, they predominantly focus on capturing observable correlations between entities. Due to the sparsity of KGs, potential semantic correlations are challenging to capture. To tackle this problem, we propose a model entitled semantic-enhanced knowledge graph completion (SE-KGC). SE-KGC effectively addresses the issue by incorporating predefined semantic patterns, enabling the capture of semantic correlations between entities and enhancing features for representation learning. To implement this approach, we employ a multi-relational graph convolution network encoder, which effectively encodes the KG. Subsequently, we utilize a scoring decoder to evaluate triplets. Experimental results demonstrate that our SE-KGC model outperforms other state-of-the-art methods in link-prediction tasks across three datasets. Specifically, compared to the baselines, SE-KGC achieved improvements of 11.7%, 1.05%, and 2.30% in terms of MRR on these three datasets. Furthermore, we present a comprehensive analysis of the contributions of different semantic patterns, and find that entities with higher connectivity play a pivotal role in effectively capturing and characterizing semantic information.

This paper focuses on the problem of traffic flow forecasting, with the aim of forecasting future traffic conditions based on historical traffic data. This problem is typically tackled by utilizing spatio-temporal graph neural networks to model the intricate spatio-temporal correlations among traffic data. Although these methods have achieved performance improvements, they often suffer from the following limitations: These methods face challenges in modeling high-order correlations between nodes. These methods overlook the interactions between nodes at different scales. To tackle these issues, in this paper, we propose a novel model named multi-scale dynamic hypergraph convolutional network (MSDHGCN) for traffic flow forecasting. Our MSDHGCN can effectively model the dynamic higher-order relationships between nodes at multiple time scales, thereby enhancing the capability for traffic forecasting. Experiments on two real-world datasets demonstrate the effectiveness of the proposed method.

Research on government service quality can help ensure the success of digital government services and has been the focus of numerous studies that proposed different frameworks and approaches. Most of the existing studies are based on traditional researcher-led methods, which struggle to capture the needs of citizens. In this paper, a citizen-feedback-based analysis framework was proposed to explore citizen demands and analyze the service quality of digital government. Citizen feedback data are a direct expression of citizens' demands, so the citizen-feedback-based framework can help to obtain more targeted management insights and improve citizen satisfaction. Efficient machine learning methods used in the framework make data collection and processing more efficient, especially for large-scale internet data. With the crawled user feedback data from the Q & A e-government portal of Luzhou, Sichuan Province, China, we conducted experiments on the proposed framework to verify its feasibility. From citizens' online feedback on Q & A services, we extracted five service quality factors: efficiency, quality, attitude, compliance, and execution of response. The analysis of five service quality factors provides some management insights, which can provide a guide for improvements in Q & A services.

With the popularity of smart terminals, such as the Internet of Things, crowdsensing is an emerging data aggregation paradigm, which plays a pivotal role in data-driven applications. There are some key issues in the development of crowdsensing such as platform security and privacy protection. As the crowdsensing is usually managed by a centralized platform, centralized management will bring various security vulnerabilities and scalability issues. To solve these issues, an effective reputation-based partition scheme (RSPC) is proposed in this article. The partition scheme calculates the optimal partition size by combining the node reputation value and divides the node into several disjoint partitions according to the node reputation value. By selecting the appropriate partition size, RSPC provides a mechanism to ensure that each partition is valid, as long as the maximum permissible threshold for the failed node is observed. At the same time, the RSPC reorganizes the network periodically to avoid partition attacks. In addition, for cross-partition transactions, this paper innovatively proposes a four-stage confirmation protocol to ensure the efficient and safe completion of cross-partition transactions. Finally, experiments show that RSPC improves scalability, low latency, and high throughput for crowdsensing.

Multi-view graph clustering (MGC) has emerged as a hot research topic due to its effectiveness. Previous works mainly focus on the fusion of similarity graphs to exploit the complementary and consistent. However, these approaches suffer from three shortcomings: 1) The distance similarity estimate by the linear Euclidean metric may be incorrect in revealing the association relationship of heterogeneous multi-view samples. 2) Most of them often weight the whole view to balance the effect of different views, which ignores the importance of local samples. 3) They fail to preserve the inherent manifold structure between views via directly applying similarity matrix fusion to obtain a consistent graph. To fill these gaps, this paper proposed a novel framework called correlation fusion with instance-level auto-weighted for MGC (CFIMGC). In particular, the Spearman distance is leveraged to effectively characterize the versatile correlation matrix. The sample-level auto-weighted fusion mechanism is presented for flexible assigning weight values to different sample-wise relations in intra-view and inter-view, through which the complementary of numerous correlation matrices is appropri-ately explored to produce a consistent correlation matrix. What's more, a probability graph is adaptively learned based on the fusion correlation matrix for constructing a clustering-friendly similarity graph. Meanwhile, an alternating iteration algorithm is devised to address the non-convex joint optimization problem. Comprehensive experiment results on seven widely used but challenging multi-view datasets demonstrate that the proposed method yields superiority in clustering performance against six state-of-the-art counterparts.