N6,2'-O-dimethyladenosine (m6Am) modification, mediated by the methyltransferase phosphorylated CTD interacting factor 1 (PCIF1), is increasingly recognized as an important layer of gene regulation. However, the role of m6Am in pain and comorbid anxiety has not yet been explored. Here, we report that persistent peripheral inflammation reduces PCIF1 levels in the mouse anterior cingulate cortex (ACC). This downregulation is triggered by decreased binding of GLI2 to the Pcif1 promoter. Knocking out or conditionally deleting Pcif1 in ACC neurons reduces the m6Am level of Gap43 mRNA, leading to the increased expression of GAP43 and pre-long-term potentiation (LTP) oversaturation, which in turn results in inflammatory pain and its comorbid anxiety-like behavior. Conversely, knocking down GAP43 suppresses the pre-LTP oversaturation caused by PCIF1 reduction, thereby alleviating inflammatory pain and comorbid anxiety. This study reveals a mechanism whereby GLI2-governed PCIF1 contributes to the modulation of inflammatory pain and comorbid anxiety through the targeting of Gap43 m6Am in the ACC.

Link prediction typically refers to the correlated neighborhood structure of nodes to predict the missing edges in graphs. Due to the incompatibility between the topological structure and node attribute in graphs, the graph neural networks (GNNs) exhibit limited performance. Previous works mainly follows a decouple-and-fuse GNNs paradigm, i.e., embedding the topological structure and node attribute in a decoupled manner and then simply fusing them together. Such methods fail to adaptively deal with the discrepancy between topological structure and node attribute. Moreover, they are often plagued by tuning the parameters between structure and attribute in the decouple-and-fuse mechanism.Thus, in this paper, we propose a Spatial Topological Discrepancy guided Adaptive-scale Graph Structure Learning (STD-AGSL) approach for link prediction. In detail, we first analyze the discrepancy between the topological structure and node attribute by designing a lightweight topological data analysis (TDA) module, generating a variable termed spatial topological discrepancy to quantify the incompatibility of graphs and examine the effect of incompatibility on the prediction performance. Moreover, we design a STD-controlled threshold function to control the update of adjacency matrices in the graph structure learning process, enabling the adaptive-scale graph structure optimization and avoiding the complex hyperparameter tuning. Finally, we feed the updated adjacency matrix into GNNs to perform the downstream link prediction task and obtain the updated node embeddings as the node attribute for the next optimization iteration. Extensive experimental results on three public benchmark datasets show that STD-AGSL significantly outperforms the vanilla GNNs in terms of MRR and Hit@100.

Histone lactylation is a novel lactate-dependent histone mark that plays an important role in regulating gene expression. However, its role in neuropathic pain remains elusive. Here, we report that Histone 3 lysine 18 lactylation (H3K18la) is a key factor in initiating and advancing neuropathic pain. Peripheral nerve injury increased the levels of H3K18la level in injured dorsal root ganglia (DRGs) of mice. This upregulation led to increased binding at the Igf2bp2 CpG island, which in turn activated Igf2bp2 transcription. Increased IGF2BP2 levels subsequently stabilized Cct2 mRNA and upregulated CCT2 protein expression. Pharmacological inhibition of H3K18la activity by injection of 2-DG or knockdown of Igf2bp2 decreased IGF2BP2 levels. This resulted in Cct2 instability in the DRG of CCI male and female mice. Concurrently, mechanical and thermal pain hypersensitivity was attenuated. In contrast, injecting sodium lactate into the uninjured DRG increased H3K18la and IGF2BP2 levels subsequently elevating CCT2 protein and inducing neuropathic pain-like behaviors. Our results reveal that histone lactylation contributes to dysfunctional gene expression in nociceptive pathways, suggesting that targeting histone lactylation could provide a new therapeutic strategy for neuropathic pain.

Zero-shot cross-domain sequential recommendation aims to transfer sequential knowledge from an interaction-rich source domain to interaction-unseen target domains without any target-domain interaction data during training. Recent large language model (LLM)-based methods have improved this setting by leveraging item-side semantic information. However, existing alignment strategies typically enforce uniform cross-domain attraction over all item pairs. Such non-selective alignment may distort the latent semantic structures and weaken the domain-specific discrimination, especially when semantically weakly related items are forced to align. To address this issue, we propose SAGERec, a similarity-aware generalization framework for zero-shot cross-domain sequen tial recommendation. SAGERec mainly comprises a Similarity-Aware Inter-domain Compactness (SIC) module that selectively aligns the semantically compatible cross-domain item pairs, and a Domain-Adaptive Generalization (DAG) module that dynamically adjusts the strength of semantic regularization according to the discrepancy between the source and target domains. In this way, the proposed framework facilitates reliable transfer while reducing noisy cross-domain attraction. Extensive experiments on multiple real-world datasets and representative sequential recommendation backbones demonstrate that SAGERec consistently improves recommendation per formance within backbone families and achieves the best overall results across zero-shot transfer settings. Further analyses indicate that, compared with existing semantic-based and generalization-based baselines, SAGERec yields more structured cross-domain alignment and a more favorable efficiency-effectiveness trade-off.

Dynamic graph contrastive learning methods commonly assume that nodes preserve stable structural roles over time. However, real-world networks often violate this assumption: nodes may change their connectivity patterns, reactivate after long inactivity, or undergo local structural reorganisation. Such heterogeneity not only weakens temporal consistency constraints but also causes volatile nodes to generate unreliable negative signals. This work shows that temporal contrasts vary substantially in reliability and should not be treated uniformly. We introduce an uncertainty-aware dynamic graph contrastive learning framework driven by temporal contrast uncertainty, which quantifies the reliability of cross-temporal contrasts directly from the graph structure. Nodes with low temporal contrast uncertainty exhibit stable structural patterns, whereas nodes with high temporal contrast uncertainty indicate volatile evolution. Leveraging temporal contrast uncertainty, the proposed framework integrates three mechanisms: (1) adaptive temporal sampling, (2) uncertainty-weighted contrastive objectives, and (3) group-level prototype calibration to correct estimation bias. We theoretically show that temporal contrast uncertainty weighting yields a favourable trade-off with respect to the mutual information lower bound, while prototype calibration provides an upper bound for controlling temporal drift. Extensive experiments on five benchmark datasets for node classification and seven datasets (including two with real-world anomaly labels) for anomaly detection demonstrate that the proposed framework consistently outperforms state-of-the-art methods, with pronounced improvements on datasets containing a high proportion of structurally volatile nodes.

Dynamic graph learning on temporal networks aims to understand the continuous evolution pattern of networks, with an important application on forecasting the future temporal network. Existing methods mainly focus on modeling the structural and temporal features, with recent research interest shifting towards considering the structural correlations between nodes through their neighbor co-occurrences. Though satisfactory performance has been achieved, there still remain several limitations: (1) the deviation of investigated scenarios from real-world applications, since most previous researches concentrate on special cases of multigraphs with abundant repeat edges; (2) the insufficient computational efficiency of modeling the structural features, since the existing neighbor co-occurrence scheme fails to consider explicit structural correlations between nodes and suffers from a time-consuming pair-wise encoding strategy; (3) the unsatisfying prediction accuracy due to inadequate modeling of temporal features, since each neighbor’s historical temporal features and the temporal domain shifting with network evolving are both neglected. To solve these issues, we first focus on the general scenarios of temporal networks without abundant repeat edges for approaching the actual applications, and propose an efficient and effective dynamic graph learning method named LightDyG. Specifically, (1) on the one hand, to increase the computational efficiency, LightDyG decouples the structural correlations between nodes and their individual substructures for fast convergence based on the analysis of existing co-occurrence mechanism, and further designs an incremental strategy for efficient structural encoding; (2) on the other hand, to improve the prediction accuracy, the temporal characteristics are considered by including both the interaction and appearance timestamps of neighbors, and a time-invariant temporal encoding strategy is designed to eliminate the temporal bias introduced by the network evolution. Extensive experiments conducted on four public temporal networks demonstrate that LightDyG outperforms the best baselines by 4.54%~11.39% and 6.06%~16.24% in terms of AP and AUC on the temporal link prediction tasks, respectively. In addition, LightDyG reduces the time cost for training and test up to 45.91% and 63.94%, respectively, and also achieves a fast convergence speed during training. The implementation of our approach is available in https://github.com/nudtzpan/LightDyG.

Inspired by the success of prompt in natural language processing, the graph prompt-based methods are proposed to solve the classification tasks under the conditions with limited instances. Recent graph prompt-based methods typically employ link prediction as the pre-training task, which brings a gap between the pre-training task and the downstream classification task, thus introducing irrelevant and noisy features to the downstream tasks. To tackle this issue, we propose a framework called Diffused Graph Prompt (Di-Graph), which consists of three major components, i.e., the diffused pre-training, a graph prompt layer, and an information bottleneck optimizer. Specifically, the diffused pre-training aims to obtain the stable node features with a diffusion process, mitigating the gap between the pre-training tasks and the downstream tasks. The graph prompt layer enhances the pre-trained model to leverage its knowledge via capturing both the structural and node features in graphs. The information bottleneck optimizer helps the model discard redundant features by retaining the minimal sufficient statistic of the input data. Extensive experimental results on five public graph datasets demonstrate that our Di-Graph model surpasses the state-of-the-art model in terms of accuracy for both node and graph classification tasks. In particular, for graph-level tasks, Di-Graph achieves an average performance gain of 6.50% over the previous best model (GraphCL) on BZR dataset. For node-level tasks, Di-Graph achieves a 2.05% improvement over the best baseline (GraphCL) on PROTEINS dataset.

Inductive link prediction on temporal networks aims to predict the future links associated with node(s) unseen in the historical timestamps. Existing methods generate the predictions mainly by learning node representation from the node/edge attributes as well as the network dynamics or by measuring the distance between nodes on the temporal network structure. However, the attribute information is unavailable in many realistic applications and the structure-aware methods highly rely on nodes' common neighbors, which are difficult to accurately detect, especially in sparse temporal networks. Thus, we propose a distance-aware learning (DEAL) approach for inductive link prediction on temporal networks. Specifically, we first design an adaptive sampling method to extract temporal adaptive walks for nodes, increasing the probability of including the common neighbors between nodes. Then, we design a dual-channel distance measuring component, which simultaneously measures the distance between nodes in the embedding space and on the dynamic graph structure for predicting future inductive edges. Extensive experiments are conducted on three public temporal network datasets, i.e., MathOverflow, AskUbuntu, and StackOverflow. The experimental results validate the superiority of DEAL over the state-of-the-art baselines in terms of accuracy, area under the ROC curve (AUC), and average precision (AP), where the improvements are especially obvious in scenarios with only limited data.

Link prediction on temporal networks aims to predict the future edges by modeling the dynamic evolution involved in the graph data. Previous methods relying on the node/edge attributes or the distance on the graph structure are not practical due to the deficiency of the attributes and the limitation of the explicit distance estimation, respectively. Moreover, the existing graph representation learning methods mostly rely on graph neural networks (GNNs), which cannot adequately take the dynamic correlations between nodes into consideration, leading to the generating of inferior node embeddings. Thus, we propose a time-aware graph (TAG) learning method for link prediction on temporal networks. We first conduct a theoretical causal analysis proving that the correlations between nodes are required to be unchanged for the temporal graph representation learning using GNNs. Then, we model the recent dynamic node correlations by designing an edge-dropping (ED) module and adopting a recent neighbor sampling (RNS) strategy so as to approximate the above condition. Besides, we also preserve the long-term stable node correlations by introducing additional self-supervisions using the contrastive learning. Comprehensive experiments were conducted on four public temporal network datasets, i.e., MathOverflow, StackOverflow, AskUbuntu, and SuperUser, demonstrate that TAG can achieve state-of-the-art performance in terms of average precision (AP) and area under the ROC curve (AUC). In addition, TAG can ensure high computational efficiency by making the temporal graph lightweight, letting it be practical in real-world applications.

Antifungal screening identified Pogostemon cablin essential oil (EO) as a potent fungicide against Colletotrichum fragariae, prompting bioassay-directed fractionation. This process identified patchoulol and pogostone as antifungal constituents. Pogostone exhibited strong activity with IC50 values of 8.5 ± 0.5, 2.34 ± 0.13, 7.74 ± 0.23, and 6.34 ± 0.85 μg mL-1 against C. fragariae, C. gloeosporioides, C. acutatum, and Botrytis cinerea, respectively, at 48 h. Due to its structural similarities to β-triketone herbicides, its phytotoxicity was assessed. It exhibited herbicidal activity against Lemna paucicostata (IC50 = 0.9 μM) comparable to atrazine (IC50 = 1.1 ± 0.23 μM), and Amaranthus palmeri in postemergence trials, showing similar efficacy to glyphosate. Despite its herbicidal effect, photosystem II disruption is unlikely, suggesting a novel mechanism. Molecular modeling also supported the idea that it may act through a distinct mode from known HPPD inhibitors. These results indicate pogostone's potential as a dual-function agrochemical with both antifungal and herbicidal properties.

Dynamic graph link prediction is widely utilized in the complex web of the real world, such as social networks, citation networks, recommendation systems, etc. Recent Transformer-based link prediction methods on dynamic graphs not only fail to model the fine-grained structures such as triangles with the vanilla Transformers in the graph serialization process, but also amplify the imbalanced distribution of graphs because of their over-estimation of high-degree nodes. To tackle these issues, we propose a Topology-aware Transformer on Dynamic Graph (TopDyG) for link prediction, consisting of a topology injected Transformer (Ti-Transformer) and a mutual information learning (Mi-Learning). % mainly consisting of two components, i.e., the topology-injected (Ti) Transformer and the mutual information (Mi) learning module. The Ti-Transformer explores the explicit structure of serialized graphs, capturing the topological features. The Mi-Learning mines the relationship between nodes by modeling the mutual information with a prior knowledge, alleviating the over-estimation of high-degree nodes when applying the Transformer-based models for the dynamic graph link prediction task. Extensive experiments on four public datasets containing both transductive and inductive settings present the superiority of our proposal. In particular, TopDyG presents an improvement of 43.27% and 28.75% against the state-of-the-art baselines in terms of NDCG and Jaccard, respectively. The advantages are especially obvious on the high-density graphs.

This study investigates the phytochemical composition and potential fungicidal and phytotoxic effects of the essential oil (EO) extracted from Piper crassinervium. Gas chromatography-mass spectrometry analysis revealed a diverse array of bioactive compounds, with myristicin (30.3%) as the predominant constituent, followed by elemicin (8.6%), epi-beta-caryophyllene (7.7%), and beta-caryophyllene (5.4%). The EO's antifungal activity was evaluated in vitro against several plant pathogens, including Colletotrichum fragariae, Colletotrichum gloeosporioides, Colletotrichum acutatum, Botrytis cinerea, and Fusarium oxysporum. A bioassay-guided fractionation approach was employed to isolate the EO's antifungal constituents, which were subsequently characterized using 1D and 2D NMR techniques. Among the isolated compounds, elemicin exhibited impressive antifungal activities against C. fragariae and B. cinerea, with IC50 values of 13.23 and 25.88 mu g/mL, respectively. Herbicidal activity was evaluated using Arabidopsis thaliana and Lemna paucicostata seed assays. Apiole and myristicin exhibited stronger inhibitory effects on A. thaliana than elemicin, with IC50 values of 68.3, 248.9, and 360.7 mu M, respectively. Apiole and myristicin also demonstrated notable inhibitory activity against L. paucicostata, with IC50 values of 166.5 and 285.1 mu M, respectively. To the best of our knowledge, this is the first study to report on the significant antifungal activity of P. crassinervium EO and its isolated pure compounds against C. fragariae and B. cinerea, as well as their moderate phytotoxic effects on A. thaliana and L. paucicostata. These findings highlight the potential of plant-derived EOs as natural alternatives to synthetic pesticides, offering promising solutions for sustainable agricultural practices and effective pest management strategies.

The gut microbiota has been linked to a number of neurological disorders. However, it is unclear whether the gut microbiota is involved in the genesis of chronic itch, a refractory condition that afflicts patients both physically and mentally. Here, we report that depletion of gut microbiota enhances tolerance to itch in mice orally administered with antibiotics (ABX) and mice free of germ. Of note, oral gavage with Bacteroides fragilis (B. fragilis), a prominent species of the genus Bacteroides with most differential change, corrected the ABX-induced itch dysfunction through its driven metabolite acetyl-l-carnitine (ALC). Mechanistically, gut microbiota or B. fragilis depletion caused a decrease in RNA N6-methyladenosine (m6A) demethylase FTO expression in the dorsal horn and a consequent increase in RNA m6A sites in Mas-related G protein-coupled receptor F (MrgprF) mRNA, leading to decreased MRGPRF protein. The downregulation of FTO was triggered by inactivation of ETS proto-oncogene 1 (ETS1), a transcription factor that binds to the Fto promoter. These findings support a gut microbe - spinal connection in modulation of itch sensation in RNA m6A epigenetic-dependent manner and highlight a critical role of ALC in linking the altered B. fragilis and itch dysfunction.

The involvement of Bicaudal D2 (BICD2), a dynein motor adaptor protein, in neuropathic pain pathogenesis remains unexplored. This study investigated the role of spinal BICD2 and its interaction with postsynaptic density protein 95 (PSD95) in neuropathic pain using Western blotting, immunofluorescence staining and reverse transcription quantitative polymerase chain reaction. We further evaluated the long-term analgesic efficacy of intrathecally (i.t.) administered Bicd2-specific antisense oligonucleotide (ASO) in male murine models of nociceptive hypersensitivity induced by spared nerve injury (SNI), cisplatin chemotherapy, and streptozotocin (STZ)-induced diabetes. Results demonstrated that BICD2 expression was selectively upregulated in the spinal dorsal horn following SNI. Both Bicd2 siRNA and Bicd2 ASO delivered via i.t. injection significantly suppressed BICD2 overexpression and attenuated mechanical/thermal hyperalgesia across all neuropathic pain etiologies (trauma, chemotherapy, diabetes). Critically, immunofluorescence staining confirmed co-expression of BICD2 and PSD95. Nerve injury-induced PSD95 elevation in the dorsal horn was abolished by Bicd2 siRNA or ASO treatment. Moreover, PSD95 inhibitor NA-1 not only alleviated SNI-induced the hyperalgesia, but also reversed mechanical/thermal hyperalgesia evoked by BICD2 overexpression. These findings reveal a novel mechanism wherein BICD2 alleviates neuropathic pain through PSD95 activation in the spinal dorsal horn, and establish the therapeutic potential of spinally targeted Bicd2 ASO for sustained pain management.

AIMS:This study aimed to test the hypothesis that nuclear factor of activated T cells 1 (NFAT1) signaling contributes to bone cancer pain by regulating interleukin (IL)-18 expression in spinal microglia. METHODS:This study was performed on male mice using a Lewis lung carcinoma-induced bone cancer pain model. Nociceptive behaviors were evaluated by measuring mechanical allodynia, thermal hyperalgesia, and spontaneous pain. Expression levels were measured via real-time quantitative polymerase chain reaction, western blotting, and immunofluorescence analysis. The effect of pharmacologic intervention of spinal NFAT1/IL-18 signaling on bone cancer pain was the primary outcome. RESULTS:NFAT1 expression was upregulated in the spinal microglia after tumor inoculation. Pharmacological inhibition of NFAT1 upregulation prevented and reversed bone cancer-related pain behaviors. In spinal microglia, NFAT1 inhibition decreased p38 MAPK phosphorylation and IL-18 production. Blocking NFAT1 signaling suppressed tumor-induced neuronal sensitization and microglial activation as well as activation of the N-methyl-D-aspartate receptor and the subsequent Ca2+-dependent signaling. CONCLUSION:Microglia NFAT1-p38 signaling contributes to bone cancer pain through IL-18-mediated central sensitization in spinal microglia. NFAT1 could be a potential target for therapeutic intervention to prevent bone cancer pain.

Dynamic link prediction aims to predict the future links on dynamic graphs, which can be applied to wide scenarios such as recommender systems and social networks on the World Wide Web. Existing methods mainly (1) focus on the in-graph learning, which cannot generalize to graphs unobserved during training; or (2) achieve the cross-graph predictions in a many-many mechanism by training on multiple graphs across various domains, which results in a large computational cost. In this paper, we propose a cross-graph dynamic link predictor named CrossDyG, which achieves the cross-graph transferability in a one-many mechanism which trains on one single source graph and test on different target graphs. Specifically, we provide causal and empirical analysis on the structural bias caused by the graph-specific structural characteristics in cross-graph predictions. Then, we conduct deconfounded training to learn the universal network evolution pattern from one single source graph during training. Finally, we apply the causal intervention to leverage the graph-specific structural characteristics of each target graph during inference. Extensive experiments conducted on three benchmark data of dynamic graphs demonstrate that CrossDyG outperforms the state-of-the-art baselines by up to 11.01% and 17.02% in terms of AP and AUC, respectively. In addition, the improvements are especially significant when training on small source graphs. The implementation of our approach is available in https://anonymous.4open.science/r/CrossDyG-8B70.

Nerve injury-induced changes in pain-associated genes contribute to genesis of neuropathic pain and comorbid anxiety. Phosphorylated CTD interacting factor-1 (PCIF1)-triggered N6, 2'-O-dimethyladenosine (m6Am) mRNA modification represents an additional layer of gene regulation. However, the role of PCIF1 in these disorders is elusive. Here, we report PCIF1 is increased in glutamatergic neurons of the hindlimb region of the primary somatosensory cortex in mouse with neuropathic pain and anxiety, but not inflammatory pain or anxiety alone. Serpine-1 mRNA-binding protein-1 (SERBP1) is identified as a PCIF1 cofactor, their complex mediates m6Am deposition onto mRNA. Blocking SERBP1-PCIF1 upregulation in glutamatergic neurons of the hindlimb region of the primary somatosensory cortex abolishes m6Am gain on maf1 homolog, negative regulator of RNA polymerase III (Maf1), elevates MAF1 protein, and mitigates neuropathic pain and anxiety. Conversely, mimicking this increase adds m6Am onto Maf1, reduces MAF1, and induces comorbidity symptoms. These findings highlight the significance of m6Am in neuropathic pain-anxiety comorbidity and identify SERBP1-PCIF1 in glutamatergic neurons of the hindlimb region of the primary somatosensory cortex as a potential therapeutic target.

Multi-modal recommenders (MRs) focus on leveraging the item modality features to facilitate user preferences modeling. Previous research mainly suffers from two limitations: (1) The pre-trained modality features are usually extracted by the encoders trained on general tasks (e.g., text classification), and thus inevitably contain the recommendation-irrelevant features. (2) Existing modality fusion mechanisms often diminish the contribution of features from weaker modalities, leading to biased fused representations. To address these challenges, we propose a novel Behavior Conditional Difussion model forMulti-Modal recommendation (BCDMM). Specifically, we first design a Behavior Multi-modal Diffusion (BMD) module to filter the recommendation-irrelevant noise within the pre-trained modality features. Then, we iteratively denoise the modality features with the guidance of user behavior signals to reconstruct the recommendation-related features. Next, we apply a Multi-modal Graph Fusion (MGF) module to explore the item modality latent structures. Moreover, we construct a modality fusion graph to capture the cross-modal complementary features for comprehensively modeling user preferences. Finally, a set of adversarial loss functions is used to balance the preservation of modalityspecific and modality-shared features. Extensive experiments on three real-world datasets demonstrate the superiority of our method. We release our code at https://github.com/fanko79/BCDMM2025.

Although a single nucleotide polymorphism for N-acetyltransferase 10 (NAT10) has been identified in patients with early-onset stroke, the role of NAT10 in ischemic injury and the related underlying mechanisms remains elusive. Here, we provide evidence that NAT10, the only known RNA N4-acetylcytidine (ac4C) modification "writer", is increased in the damaged cortex of patients with acute ischemic stroke and the peri-infarct cortex of mice subjected to photothrombotic (PT) stroke. Pharmacological inhibition of NAT10 with remodelin on Days 3-7 post-stroke or astrocytic depletion of NAT10 via targeted virus attenuates ischemia-induced infarction and improves functional recovery in PT mice. Mechanistically, NAT10 enhances ac4C acetylation of the inflammatory cytokine tissue inhibitor of metalloproteinase 1 (Timp1) mRNA transcript, which increases TIMP1 expression and results in the accumulation of microtubule-associated protein 1 light chain 3 (LC3) and progression of astrocyte autophagy. These findings demonstrate that NAT10 regulates astrocyte autophagy by targeting Timp1 ac4C after stroke. This study highlights the critical role of ac4C in the regulation of astrocyte autophagy and proposes a promising strategy to improve post-stroke outcomes via NAT10 inhibition.

Multi-modal recommendation has received significant attention in recent years. It typically leverages the item modality features to enhance the item semantics and then incorporates these features into the final user/item representations. Existing studies primarily focus on graph-based multi-modal recommendation. However, they often neglect the impact of noisy interaction, and this impact tends to be amplified as neighborhood messages are aggregated. Further, we argue that existing methods fail to distinguish the relative importance of modality-specific graph embeddings in the fusion process. To address these issues, we propose a novel Denoising Attentive Hypergraph Learning method for Multi-Modal Recommendation (DAHM). Specifically, we first de vise a Di richlet degree-sensitive method to prune the potential noisy edges. Then, we introduce a local user interest graph encoder and a global item attribute hypergraph encoder to capture the local user preferences and the global similarities among the item modality attributes, respectively. Next, a projection-enhancement hypergraph attention mechanism is applied to adaptively allocate the fusion weights to modality-specific graph embeddings. Extensive experiments on three real-world datasets verify the effectiveness of DAHM, presenting an average improvement of 9.55% in terms of Recall@20 against the best baseline. We release our code at https://github.com/fanko79/DAHM2025.