Peng Wu’s research while affiliated with Guangzhou University of Chinese Medicine and other places

... Large Language Models (LLMs) are getting increasingly accepted as the standard in many industry applications (Chkirbene et al., 2024), achieving state-of-the-art performance in tasks such as sentiment analysis, machine translation, and claim verification (Miah et al., 2024;Zhang et al., 2023;Xu et al., 2024). However, their dominance has also raised concerns-LLMs require large-scale computational resources, leaving behind a large carbon footprint (Faiz et al., 2023) which makes them overkill for many real-world applications. ...

Reference:

Non-Euclidean Hierarchical Representational Learning using Hyperbolic Graph Neural Networks for Environmental Claim Detection

... Previous literature has reported that the anti-inflammatory effects of ponicidin are associated with the inhibition of macrophages and the production of pro-inflammatory cytokines (22). Our findings indicate that ponicidin can effectively suppress the infiltration of inflammatory cells and the release of proinflammatory cytokines in the hind paw. ...

Reference:

The effect of ponicidin on CFA-induced chronic inflammatory pain and its mechanism based on network pharmacology and molecular docking

... Primary peritoneal macrophages were prepared from rats as described below [22]. Briefly, rats were killed by carbon dioxide, and 10 mL of pre-cooled DMEM F-12 medium was injected intraperitoneally. ...

Reference:

Oxymatrine Attenuates Ulcerative Colitis through Inhibiting Pyroptosis Mediated by the NLRP3 Inflammasome

... Previous works abandon the attentive filtering of structured knowledge by leveraging mean pooling to avoid the nested attention (Dun et al., 2021;Tseng et al., 2022). Instead, we adopt a modulation mechanism based on product & max pooling (Ouyang et al., 2022) to attentively filter the related structured knowledge with lower computational complexity. Specifically, given an entity en i and its neighbors N (en i ), the structured knowledge is filtered as: ...

Reference:

COOL: Comprehensive Knowledge Enhanced Prompt Learning for Domain Adaptive Few-shot Fake News Detection

... Zheng et al. [36] introduced the JORA model that uses an inductive GCN for low-dimensional user representations while maintaining essential features. JORA is optimized via representation learning to preserve similarities within networks and alignment learning to project and align across networks using hard and attention-based soft alignment mechanisms. ...

Reference:

User Identity Linkage on Social Networks: A Review of Modern Techniques and Applications

... Embedding learning based on heterogeneous networks can integrate multi-dimensional heterogeneous information in the network and generate node vectors with rich semantics. The supervised embedding learning models can optimize the parameters in the model with the help of supervision information, and extract the features in the network to encode the nodes [1][2][3]. However, in practice, because it is difficult to label large-scale network data, most heterogeneous networks do not contain supervision information. ...

Reference:

Unsupervised Embedding Learning for Large-Scale Heterogeneous Networks Based on Metapath Graph Sampling

... Furthermore, the elevation of CTSB persisted up to 24 h post-PCI. 58 Numerous studies in the field of animal experimental research have demonstrated a significant upregulation of CTSB expression in the serum and heart of rats with isoproterenol (ISO)-induced MI. [59][60][61] This upregulation of CTSB can activate downstream proinflammatory mediators, thereby playing a role in the pathophysiological processes of MI. 61 Additionally, the CTSB inhibition has been shown to effectively mitigate cardiac dysfunction, limit infarct size and reduce inflammatory invasion in Sprague-Dawley rats following left anterior descending coronary artery ligation, achieved through the inhibition of NLRP3 activation. 58 From the information presented above, it can be inferred that CTSB significantly impacts MI, potentially participating in critical pathological mechanisms like plaque rupture and post-MI myocardial remodelling. ...

Reference:

Cathepsin B in cardiovascular disease: Underlying mechanisms and therapeutic strategies

... The main cause of BBB dysfunction is not only inflammation, but also oxidative stress. Nuclear factor E2-related factor 2 (Nrf2) is a protective transcription factor of antioxidant stress [66]. Nrf2 transfers to the nucleus to bind with antioxidant response elements (AREs) and promote the expression of phase II detoxifying and antioxidant enzymes, including heme oxygenase-1 (HO-1), among others [67]. ...

Reference:

GSK3β Substrate-competitive Inhibitors Regulate the gut Homeostasis and Barrier Function to Inhibit Neuroinflammation in Scopolamine-induced Alzheimer’s Disease Model Mice

... Ramezani et al. [109] propose a probabilistic generative model that maps the interrelationships between chain edges and cascade processes with a coupling matrix decomposition to infer the underlying social network structure and information dissemination. Considering the temporal, Xu et al. [110] incorporate the time feature into the proximity matrix and utilize hierarchical clustering methods to classify multi-scale information, which improves the model's performance. Liu et al. [111] use Markov to analyze the information dissemination process, which considers an adaptive network's information state and network topology. ...

Reference:

A Survey of Information Dissemination Model, Datasets, and Insight

... Deep Learning-Based Methods: Distinguished by their use of deep learning, these methods (43,44) leverage the advantages of deep neural networks to extract high-order nonlinear relationships from graph data. ...

Reference:

EEG-based epilepsy detection with graph correlation analysis