Distantly Supervised Biomedical Relation Extraction Via Negative Learning and Noisy Student Self-Training.

Biomedical relation extraction aims to identify underlying relationships among entities, such as gene associations and drug interactions, within biomedical texts. Despite advancements in relation extraction in general knowledge domains, the scarcity of labeled training data remains a significant challenge in the biomedical field. This paper provides a novel approach for biomedical relation extraction that leverages a noisy student self-training strategy combined with negative learning. This method addresses the challenge of data insufficiency by utilizing distantly supervised data to generate high-quality labeled samples. Negative learning, as opposed to traditional positive learning, offers a more robust mechanism to discern and relabel noisy samples, preventing model overfitting. The integration of these techniques ensures enhanced noise reduction and relabeling capabilities, leading to improved performance even with noisy datasets. Experimental results demonstrate the effectiveness of the proposed framework in mitigating the impact of noisy data and outperforming existing benchmarks.

A Radical Activation Strategy for Versatile and Stereoselective N-Glycosylation.

Previous N-glycosylation approaches have predominately involved acidic conditions, facing challenges of low stereoselectivity and limited scope. Herein, we introduce a radical activation strategy that enables versatile and stereoselective N-glycosylation using readily accessible glycosyl sulfinate as a donor under basic conditions and exhibits exceptional tolerance towards various N-aglycones containing alkyl, aryl, heteroaryl and nucleobase functionalities. Preliminary mechanistic studies indicate a pivotal role of iodide, which orchestrates the formation of a glycosyl radical from the glycosyl sulfinate and subsequent generation of the key intermediate, a configurationally well-defined glycosyl iodide, which is subsequently attacked by an N-aglycone in a stereospecific SN2 manner to give the desired N-glycosides. An alternative route involving the coupling of a glycosyl radical and a nitrogen-centered radical is also proposed, affording the exclusive 1,2-trans product. This novel approach promises to broaden the synthetic landscape of N-glycosides, offering a powerful tool for the construction of complex glycosidic structures under mild conditions.

Consistent time allocation fraction to vegetation green-up versus senescence across northern ecosystems despite recent climate change.

Extended growing season lengths under climatic warming suggest increased time for plant growth. However, research has focused on climatic impacts to the timing or duration of distinct phenological events. Comparatively little is known about impacts to the relative time allocation to distinct phenological events, for example, the proportion of time dedicated to leaf growth versus senescence. We use multiple satellite and ground-based observations to show that, despite recent climate change during 2001 to 2020, the ratio of time allocated to vegetation green-up over senescence has remained stable [1.27 (± 0.92)] across more than 83% of northern ecosystems. This stability is independent of changes in growing season lengths and is caused by widespread positive relationships among vegetation phenological events; longer vegetation green-up results in longer vegetation senescence. These empirical observations were also partly reproduced by 13 dynamic global vegetation models. Our work demonstrates an intrinsic biotic control to vegetation phenology that could explain the timing of vegetation senescence under climate change.

Chromosome-level genome assembly provides insights into the genetic diversity, evolution, and flower development of Prunus conradinae.

Prunus conradinae, a valuable flowering cherry belonging to the Rosaceae family subgenus Cerasus and endemic to China, has high economic and ornamental value. However, a high-quality P. conradinae genome is unavailable, which hinders our understanding of its genetic relationships and phylogenesis, and ultimately, the possibility of mining of key genes for important traits. Herein, we have successfully assembled a chromosome-scale P. conradinae genome, identifying 31,134 protein-coding genes, with 98.22% of them functionally annotated. Furthermore, we determined that repetitive sequences constitute 46.23% of the genome. Structural variation detection revealed some syntenic regions, inversions, translocations, and duplications, highlighting the genetic diversity and complexity of Cerasus. Phylogenetic analysis demonstrated that P. conradinae is most closely related to P. campanulata, from which it diverged ~ 19.1 million years ago (Mya). P. avium diverged earlier than P. cerasus and P. conradinae. Similar to the other Prunus species, P. conradinae underwent a common whole-genome duplication event at ~ 138.60 Mya. Furthermore, 79 MADS-box members were identified in P. conradinae, accompanied by the expansion of the SHORT VEGETATIVE PHASE subfamily. Our findings shed light on the complex genetic relationships, and genome evolution of P. conradinae and will facilitate research on the molecular breeding and functions of key genes related to important horticultural and economic characteristics of subgenus Cerasus.

Heterogeneous graph convolutional network for multi-view semi-supervised classification.

This paper proposes a novel approach to semantic representation learning from multi-view datasets, distinct from most existing methodologies which typically handle single-view data individually, maintaining a shared semantic link across the multi-view data via a unified optimization process. Notably, even recent advancements, such as Co-GCN, continue to treat each view as an independent graph, subsequently aggregating the respective GCN representations to form output representations, which ignores the complex semantic interactions among heterogeneous data. To address the issue, we design a unified framework to connect multi-view data with heterogeneous graphs. Specifically, our study envisions multi-view data as a heterogeneous graph composed of shared isomorphic nodes and multi-type edges, wherein the same nodes are shared across different views, but each specific view possesses its own unique edge type. This perspective motivates us to utilize the heterogeneous graph convolutional network (HGCN) to extract semantic representations from multi-view data for semi-supervised classification tasks. To the best of our knowledge, this is an early attempt to transfigure multi-view data into a heterogeneous graph within the realm of multi-view semi-supervised learning. In our approach, the original input of the HGCN is composed of concatenated multi-view matrices, and its convolutional operator (the graph Laplacian matrix) is adaptively learned from multi-type edges in a data-driven fashion. After rigorous experimentation on eight public datasets, our proposed method, hereafter referred to as HGCN-MVSC, demonstrated encouraging superiority over several state-of-the-art competitors for semi-supervised classification tasks.

Noninvasive diagnosis of interstitial fibrosis in chronic kidney disease: a systematic review and meta-analysis.

RATIONALE AND OBJECTIVES: Researchers have delved into noninvasive diagnostic methods of renal fibrosis (RF) in chronic kidney disease, including ultrasound (US), magnetic resonance imaging (MRI), and radiomics. However, the value of these diagnostic methods in the noninvasive diagnosis of RF remains contentious. Consequently, the present study aimed to systematically delineate the accuracy of the noninvasive diagnosis of RF. MATERIALS AND METHODS: A systematic search covering PubMed, Embase, Cochrane Library, and Web of Science databases for all data available up to 28 July 2023 was conducted for eligible studies. RESULTS: We included 21 studies covering 4885 participants. Among them, nine studies utilized US as a noninvasive diagnostic method, eight studies used MRI, and four articles employed radiomics. The sensitivity and specificity of US for detecting RF were 0.81 (95% CI: 0.76-0.86) and 0.79 (95% CI: 0.72-0.84). The sensitivity and specificity of MRI were 0.77 (95% CI: 0.70-0.83) and 0.92 (95% CI: 0.85-0.96). The sensitivity and specificity of radiomics were 0.69 (95% CI: 0.59-0.77) and 0.78 (95% CI: 0.68-0.85). CONCLUSIONS: The current early noninvasive diagnostic methods for RF include US, MRI, and radiomics. However, this study demonstrates that US has a higher sensitivity for the detection of RF compared to MRI. Compared to US, radiomics studies based on US did not show superior advantages. Therefore, challenges still exist in the current radiomics approaches for diagnosing RF, and further exploration of optimized artificial intelligence (AI) algorithms and technologies is needed.

Different frequencies of water deficit irrigation treatments improve fruit quality of Zitian seedless grapes under on-tree storage.

In this study, we assessed the impact of varied water deficit irrigation frequencies (T1: 2.5 L/4 days; T2: 5 L/8 days; CK: 5 L/4 days) on Zitian Seedless grapes from veraison to post-ripening. Notably, total soluble solids increased during on-tree storage compared to at maturity, while total anthocyanin content decreased, particularly in CK (60.16%), T1 (62.35%), and less in T2 (50.54%). Glucose and fructose levels increased significantly in T1 and T2, more so in T2, but slightly declined in CK. Tartaric acid content increased by 41.42% in T2. Moreover, compared to regular irrigation, water deficit treatments enhanced phenolic metabolites and volatile compounds, including chlorogenic acid, various flavonoids, viniferin, hexanal, 2-nonenal, 2-hexen-1-ol, (E)-, 3-hydroxy-dodecanoic acid, and 1-hexanol, etc. Overall, the T2 treatment outperformed T1 and CK in maintaining grape quality. This study reveals that combining on-tree storage with water deficit irrigation not only improves grape quality but also water efficiency.

UMCGL: Universal Multi-View Consensus Graph Learning With Consistency and Diversity.

Existing multi-view graph learning methods often rely on consistent information for similar nodes within and across views, however they may lack adaptability when facing diversity challenges from noise, varied views, and complex data distributions. These challenges can be mainly categorized into: 1) View-specific diversity within intra-view from noise and incomplete information; 2) Cross-view diversity within inter-view caused by various latent semantics; 3) Cross-group diversity within inter-group due to data distribution differences. To this end, we propose a universal multi-view consensus graph learning framework that considers both original and generative graphs to balance consistency and diversity. Specifically, the proposed framework can be divided into the following four modules: i) Multi-channel graph module to extract principal node information, ensuring view-specific and cross-view consistency while mitigating view-specific and cross-view diversity within original graphs; ii) Generative module to produce cleaner and more realistic graphs, enriching graph structure while maintaining view-specific consistency and suppressing view-specific diversity; iii) Contrastive module to collaborate on generative semantics to facilitate cross-view consistency and reducing cross-view diversity within generative graphs; iv) Consensus graph module to consolidate learning a consensual graph, pursuing cross-group consistency and cross-group diversity. Extensive experimental results on real-world datasets demonstrate its effectiveness and superiority.

Attributed Multi-Order Graph Convolutional Network for Heterogeneous Graphs.

Heterogeneous graph neural networks play a crucial role in discovering discriminative node embeddings and relations from multi-relational networks. One of the key challenges in heterogeneous graph learning lies in designing learnable meta-paths, which significantly impact the quality of learned embeddings. In this paper, we propose an Attributed Multi-Order Graph Convolutional Network (AMOGCN), which automatically explores meta-paths that involve multi-hop neighbors by aggregating multi-order adjacency matrices. The proposed model first constructs different orders of adjacency matrices from manually designed node connections. Next, AMOGCN fuses these various orders of adjacency matrices to create an intact multi-order adjacency matrix. This process is supervised by the node semantic information, which is extracted from the node homophily evaluated by attributes. Eventually, we employ a one-layer simplifying graph convolutional network with the learned multi-order adjacency matrix, which is equivalent to the cross-hop node information propagation with multi-layer graph neural networks. Substantial experiments reveal that AMOGCN achieves superior semi-supervised classification performance compared with state-of-the-art competitors.

Protective role of arachidonic acid against diabetic myocardial ischemic injury: a translational study of pigs, rats, and humans.

AIM: Patients with diabetes mellitus have poor prognosis after myocardial ischemic injury. However, the mechanism is unclear and there are no related therapies. We aimed to identify regulators of diabetic myocardial ischemic injury. METHODS AND RESULTS: Mass spectrometry-based, non-targeted metabolomic approach was used to profile coronary sinus blood from diabetic and non-diabetic Bama-mini pigs at 0.5-h post coronary artery ligation. Six metabolites had a |log2 (Fold Change)|> 1.3. Among them, the most changed is arachidonic acid (AA), levels of which were 32 times lower in diabetic pigs than in non-diabetic pigs. The AA-derived products, PGI2 and 6-keto-PGF1α, were also significantly reduced. AA treatment of cultured cardiomyocytes protected against cell death by 30% at 48 h of high glucose and oxygen deprivation, which coincided with increased mitophagic activity (as indicated by increased LC3II/LC3I, decreased p62 and increased parkin & PINK1), improved mitochondrial renewal (upregulation of Drp1 and FIS1), reduced ROS generation and increased ATP production. These cardioprotective effects were abolished by PINK1(a crucial mitophagy protein) knockdown or the autophagy inhibitor 3-Methyladenine. The protective effect of AA was also inhibited by indomethacin and Cay10441, a prostacyclin receptor antagonist. Furthermore, diabetic Sprague Dawley rats were subjected to coronary ligation for 40 min and AA treatment (10 mg/day per animal gavaged) decreased myocardial infarct size, cell apoptosis index, inflammatory cytokines and improved heart function. Scanning electron microscopy showed more intact mitochondria in the border zone of infarcted myocardium in AA treated rats. Lastly, diabetic patients after myocardial infarction had lower plasma levels of AA and 6-keto-PGF1α and reduced cardiac ejection fraction, compared with non-diabetic patients after myocardial infarction. Plasma AA level was inversely correlated with fasting blood glucose. CONCLUSIONS: AA protects against diabetic ischemic myocardial damage by promoting mitochondrial autophagy and renewal, which is related to AA derived PGI2 signaling. AA may represent a new strategy to treat diabetic myocardial ischemic injury.

Assessing the effectiveness of a residential-scale detention tank operated in a multi-objective approach using SWMM.

The volume capture ratio of annual rainfall (VCRAR) of low-impact development measures is significantly influenced by its operating characteristics, particularly for residential stormwater detention tanks (SWDTs). The multi-objective operation strategy of SWDTs, encompassing toilet flushing (TF), green space irrigation (GSI), combined TF and GSI (TF-GSI), and peak flow reduction (PFR) rate, were compared using a case study in Beijing based on the stormwater management model. The findings indicate that the VCRAR for TF, GSI, and TF-GSI rainwater harvesting targets was 89.05, 77.16, and 91.21%, respectively. The operating scheme and return periods have a significant impact on the PFR rate's effectiveness. When the return period was lower than 10 years, the SWDT does not reach its maximum storage capacity, and the PFR rate was increased with increasing the return period: the PFR rate was 71.47% when the design return period was 10 years. It will also produce the phenomena of water inrush, and the overflow volume will grow rapidly when the SWDT reaches its maximum storage capacity. Hence, the operation of SWDTs may be integrated with real-time control to optimize the VCRAR for rainwater reuse and flood migration, thereby enhancing the volume utilization efficiency of SWDTs.

Effects of expanding imports on urban manufacturing employment: Evidence from China.

Full employment is important to promote the high-quality development of the urban economy. Using urban-level data on China from 2004 to 2018, we analyse the effects and mechanism of expanding imports on urban manufacturing employment. We use the Guiding Opinions on Strengthening Import to Promote Balanced Development of Foreign Trade issued by the China State Council in 2012 as a natural experiment to solve the endogeneity problem. We find that expanding imports significantly increases urban manufacturing employment. This conclusion is still robust after a series of robustness tests. Further mechanism tests reveal that productivity improvements and upgrades to product quality from expanding imports can explain increased urban manufacturing employment. The results of the heterogeneity analysis show that expanding imports promote manufacturing employment in large and medium-sized cities but not small cities. Expanding imports increases employment in manufacturing in cities in different regions, with the largest effects on eastern cities, the second largest effects on western cities, and the smallest effects on central cities. These results suggest that expanding imports is an effective channel for increasing employment.

Redefining Esophagectomy: The Manual Layered Insertion Method That May Reduce Anastomotic Leakage.

INTRODUCTION: Anastomotic leakage post-esophagectomy remains a significant challenge. Despite the use of both mechanical and manual anastomosis, leakage rates remain high. This study evaluated the effectiveness of the manual layered insertion anastomosis technique in addressing this issue. METHODS: A retrospective analysis was conducted on patients who underwent this technique from September 2020 to December 2021. The process involved thoracoscopic release of the esophagus, mediastinal lymph node dissection, laparoscopic stomach release, and its transformation into a tube. The latter was then guided to the neck for anastomosis. The posterior anastomotic wall was reshaped in the neck first for optimal insertion, followed by layered suturing with the gastric conduit. The anterior wall was subsequently sutured and repositioned into the chest. RESULTS: The study included 56 patients (51 men, five women, mean age 65.4 y), with nine having undergone neoadjuvant therapy. All received minimally invasive esophagectomy. Average intraoperative blood loss was 79.8 mL, operation time averaged 331 min, and feeding resumed after an average of 6.3 d. No anastomotic leakages were reported, with reduced incidences of anastomotic stenosis and gastric acid reflux compared to previous studies. CONCLUSIONS: The manual layered insertion anastomosis technique may reduce anastomotic leakage and associated complications, improving the efficacy of esophagectomy, which may improve postoperative results and patient quality of life, suggesting the method's potential suitability for wider clinical application.

Aspertaichamide a, a novel cytotoxic prenylated indole alkaloid possessing a bicyclo[2.2.2]diazaoctane framework from a marine algal-derived endophytic fungus aspergillus taichungensis 299.

Filamentous fungi belonging to the genus Aspergillus are prodigious producers of alkaloids, particularly prenylated indole alkaloids, that often exhibit structurally diversified skeletons and potent biological activities. In this study, five prenylated indole alkaloids possessing a bicyclo[2.2.2]diazaoctane core ring system, including a novel derivative, namely aspertaichamide A (1), as well as four known compounds, (+)-stephacidin A (2), sclerotiamide (3), (-)-versicolamide B (4), and (+)-versicolamide B (5), were isolated and identified from A. taichungensis 299, an endophytic fungus obtained from the marine red alga Gelidium amansii. The chemical structures of the compounds were elucidated by comprehensive NMR and HRESIMS spectroscopic analyses. In addition to the previously reported prenylated indole alkaloids, aspertaichamide A (1) was characterized as having an unusual ring structure with the fusion of a 3-pyrrolidone dimethylbenzopyran to the bicyclo[2.2.2]diazaoctane moiety, which was rare in these kinds of compounds. The absolute configuration of 1 was determined by TDDFT-ECD calculations. In vitro cytotoxic assays revealed that the novel compound 1 possessed selective cytotoxic activity against five human tumor cell lines (A549, HeLa, HepG2, HCT-116, and AGS), with IC50 values of 1.7-48.5 µM. Most importantly, compound 1 decreased the viability of AGS cells in a concentration-dependent manner with an IC50 value of 1.7 µM. Further studies indicated that 1 may induce AGS cells programmed cell death via the apoptotic pathway.

Revisiting multi-view learning: A perspective of implicitly heterogeneous Graph Convolutional Network.

Graph Convolutional Network (GCN) has become a hotspot in graph-based machine learning due to its powerful graph processing capability. Most of the existing GCN-based approaches are designed for single-view data. In numerous practical scenarios, data is expressed through multiple views, rather than a single view. The ability of GCN to model homogeneous graphs is indisputable, while it is insufficient in facing the heterophily property of multi-view data. In this paper, we revisit multi-view learning to propose an implicit heterogeneous graph convolutional network that efficiently captures the heterogeneity of multi-view data while exploiting the powerful feature aggregation capability of GCN. We automatically assign optimal importance to each view when constructing the meta-path graph. High-order cross-view meta-paths are explored based on the obtained graph, and a series of graph matrices are generated. Combining graph matrices with learnable global feature representation to obtain heterogeneous graph embeddings at various levels. Finally, in order to effectively utilize both local and global information, we introduce a graph-level attention mechanism at the meta-path level that allocates private information to each node individually. Extensive experimental results convincingly support the superior performance of the proposed method compared to other state-of-the-art approaches.

RhoA suppresses pseudorabies virus replication in vitro.

The porcine pseudorabies virus (PRV) is one of the most devastating pathogens and brings great economic losses to the swine industry worldwide. Viruses are intracellular parasites that have evolved numerous strategies to subvert and utilize different host processes for their life cycle. Among the different systems of the host cell, the cytoskeleton is one of the most important which not only facilitate viral invasion and spread into neighboring cells, but also help viruses to evade the host immune system. RhoA is a key regulator of cytoskeleton system that may participate in virus infection. In this study, we characterized the function of RhoA in the PRV replication by chemical drugs treatment, gene knockdown and gene over-expression strategy. Inhibition of RhoA by specific inhibitor and gene knockdown promoted PRV proliferation. On the contrary, overexpression of RhoA or activation of RhoA by chemical drug inhibited PRV infection. Besides, our data demonstrated that PRV infection induced the disruption of actin stress fiber, which was consistent with previous report. In turn, the actin specific inhibitor cytochalasin D markedly disrupted the normal fibrous structure of intracellular actin cytoskeleton and decreased the PRV replication, suggesting that actin cytoskeleton polymerization contributed to PRV replication in vitro. In summary, our data displayed that RhoA was a host restriction factor that inhibited PRV replication, which may deepen our understanding the pathogenesis of PRV and provide further insight into the prevention of PRV infection and the development of anti-viral drugs.

Development and validation of prediction models for poor sleep quality among older adults in the post-COVID-19 pandemic era.

BACKGROUND: Corona Virus Disease 2019 (COVID-19) has a significant impact on sleep quality. However, the effects on sleep quality in the post-COVID-19 pandemic era remain unclear, and there is a lack of a screening tool for Chinese older adults. This study aimed to understand the prevalence of poor sleep quality and determine sensitive variables to develop an effective prediction model for screening sleep problems during infectious diseases outbreaks. MATERIALS AND METHODS: The Peking University Health Cohort included 10,156 participants enrolled from April to May 2023. The Pittsburgh Sleep Quality Index (PSQI) scale was used to assess sleep quality. The data were randomly divided into a training-testing cohort (n = 7109, 70%) and an independent validation cohort (n = 3027, 30%). Five prediction models with 10-fold cross validation including the Least Absolute Shrinkage and Selection Operator (LASSO), Stochastic Volatility Model (SVM), Random Forest (RF), Artificial Neural Network (ANN), and XGBoost model based on the area under curve (AUC) were used to develop and validate predictors. RESULTS: The prevalence of poor sleep quality (PSQI >7) was 30.69% (3117/10,156). Among the generated models, the LASSO model outperformed SVM (AUC 0.579), RF (AUC 0.626), ANN (AUC 0.615) and XGBoost (AUC 0.606), with an AUC of 0.7. Finally, a total of 12 variables related to sleep quality were used as parameters in the prediction models. These variables included age, gender, ethnicity, educational level, residence, marital status, history of chronic diseases, SARS-CoV-2 infection, COVID-19 vaccination, social support, depressive symptoms, and cognitive impairment among older adults during the post-COVID-19 pandemic. The nomogram illustrated that depressive symptoms contributed the most to the prediction of poor sleep quality, followed by age and residence. CONCLUSIONS: This nomogram, based on twelve-variable, could potentially serve as a practical and reliable tool for early identification of poor sleep quality among older adults during the post-pandemic period.

The poor sleep quality (PSQI >7) was still prevalent among older adults during the post-COVID-19 pandemic.The LASSO model was the best model to predict poor sleep quality among older adults, compared with SVM, RF, ANN and XGBoost.This prediction model, based on twelve variables, may potentially serve as a practical and reliable tool for the early identification of poor sleep quality among older adults during the post-pandemic period.

Carbon-doped boron nitride nanosheets as an efficient metal-free catalyst for the selective oxidation of H2S.

We report the first use of carbon-doped boron nitride (BCN) for H2S-selective catalytic oxidation. The obtained carbon-doped BN with an ultrathin layer structure exhibits outstanding H2S elimination and high S yield. In particular, BN doped carbon nanosheets display better catalytic performance than traditional catalysts, such as iron- and carbon-based catalysts. The findings of the present work shed a new light on metal-free catalysts for efficient catalytic removal of toxic H2S.

An early warning signal for grassland degradation on the Qinghai-Tibetan Plateau.

Intense grazing may lead to grassland degradation on the Qinghai-Tibetan Plateau, but it is difficult to predict where this will occur and to quantify it. Based on a process-based ecosystem model, we define a productivity-based stocking rate threshold that induces extreme grassland degradation to assess whether and where the current grazing activity in the region is sustainable. We find that the current stocking rate is below the threshold in ~80% of grassland areas, but in 55% of these grasslands the stocking rate exceeds half the threshold. According to our model projections, positive effects of climate change including elevated CO2 can partly offset negative effects of grazing across nearly 70% of grasslands on the Plateau, but only in areas below the stocking rate threshold. Our analysis suggests that stocking rate that does not exceed 60% (within 50% to 70%) of the threshold may balance human demands with grassland protection in the face of climate change.

Learnable Graph Convolutional Network With Semisupervised Graph Information Bottleneck.

Graph convolutional network (GCN) has gained widespread attention in semisupervised classification tasks. Recent studies show that GCN-based methods have achieved decent performance in numerous fields. However, most of the existing methods generally adopted a fixed graph that cannot dynamically capture both local and global relationships. This is because the hidden and important relationships may not be directed exhibited in the fixed structure, causing the degraded performance of semisupervised classification tasks. Moreover, the missing and noisy data yielded by the fixed graph may result in wrong connections, thereby disturbing the representation learning process. To cope with these issues, this article proposes a learnable GCN-based framework, aiming to obtain the optimal graph structures by jointly integrating graph learning and feature propagation in a unified network. Besides, to capture the optimal graph representations, this article designs dual-GCN-based meta-channels to simultaneously explore local and global relations during the training process. To minimize the interference of the noisy data, a semisupervised graph information bottleneck (SGIB) is introduced to conduct the graph structural learning (GSL) for acquiring the minimal sufficient representations. Concretely, SGIB aims to maximize the mutual information of both the same and different meta-channels by designing the constraints between them, thereby improving the node classification performance in the downstream tasks. Extensive experimental results on real-world datasets demonstrate the robustness of the proposed model, which outperforms state-of-the-art methods with fixed-structure graphs.