Topological Similarity Research Articles

Accurate prediction of drug-target interactions in Chinese and western medicine by the CWI-DTI model

Accurate prediction of drug-target interactions (DTIs) is crucial for advancing drug discovery and repurposing. Computational methods have significantly improved the efficiency of experimental predictions for drug-target interactions in Western medicine. However, accurately predicting the complex relationships between Chinese medicine ingredients and targets remains a formidable challenge due to the vast number and high heterogeneity of these ingredients. In this study, we introduce the CWI-DTI method, which achieves high-accuracy prediction of DTIs using a large dataset of interactive relationships of drug ingredients or candidate targets. Moreover, we present a novel dataset to evaluate the prediction accuracy of both Chinese and Western medicine. Through meticulous collection and preprocessing of data on ingredients and targets, we employ an innovative autoencoder framework to fuse multiple drug (target) topological similarity matrices. Additionally, we employ denoising blocks, sparse blocks, and stacked blocks to extract crucial features from the similarity matrix, reducing noise and enhancing accuracy across diverse datasets. Our results indicate that the CWI-DTI model shows improved performance compared to several existing state-of-the-art methods on the datasets tested in both Western and Chinese medicine databases. The findings of this study hold immense promise for advancing DTI prediction in Chinese and Western medicine, thus fostering more efficient drug discovery and repurposing endeavors. Our model is available at https://github.com/WANG-BIN-LAB/CWIDTI.

The topology and geometry of neural representations

A central question for neuroscience is how to characterize brain representations of perceptual and cognitive content. An ideal characterization should distinguish different functional regions with robustness to noise and idiosyncrasies of individual brains that do not correspond to computational differences. Previous studies have characterized brain representations by their representational geometry, which is defined by the representational dissimilarity matrix (RDM), a summary statistic that abstracts from the roles of individual neurons (or responses channels) and characterizes the discriminability of stimuli. Here, we explore a further step of abstraction: from the geometry to the topology of brain representations. We propose topological representational similarity analysis, an extension of representational similarity analysis that uses a family of geotopological summary statistics that generalizes the RDM to characterize the topology while de-emphasizing the geometry. We evaluate this family of statistics in terms of the sensitivity and specificity for model selection using both simulations and functional MRI (fMRI) data. In the simulations, the ground truth is a data-generating layer representation in a neural network model and the models are the same and other layers in different model instances (trained from different random seeds). In fMRI, the ground truth is a visual area and the models are the same and other areas measured in different subjects. Results show that topology-sensitive characterizations of population codes are robust to noise and interindividual variability and maintain excellent sensitivity to the unique representational signatures of different neural network layers and brain regions.

Bioinformatic identification of monkeypox virus phylogenetic gene trees that are representative of its whole-genome phylogenetic tree.

Phylogenetic analysis based on whole-genome sequences is the gold standard for monkeypox virus (MPXV) phylogeny. However, genomic epidemiology capability and capacity are lacking or limited in resource poor countries of sub-Saharan Africa. Therefore, these make real-time genome surveillance of MPXV virtually impossible. We hypothesized that phylogenetic analysis based on single, conserved genes will produce phylogenetic tree topology consistent with MPXV whole-genome phylogeny, thus serving as a reliable proxy to phylogenomic analysis. In this study, we analyzed 62 conserved MPXV genes and showed that Bayesian phylogenetic analysis based on five genes (OPG 066/E4L, OPG068/E6R, OPG079/I3L, OPG145/A18R, and OPG150/A23R) generated phylogenetic trees with 72.2-96.3% topology similarity index to the reference phylogenomic tree topology. Our results showed that phylogenetic analysis of the identified five genes singly or in combination can serve as surrogate for whole-genome phylogenetic analysis, and thus obviates the need for whole-genome sequencing and phylogenomic analysis in regions where genomic epidemiology competence and capacity are lacking or unavailable. This study is relevant to evolution and genome surveillance of MPXV in resource limited countries.

Gene co-expression networks reveal sex-biased differences in musculoskeletal ageing.

Aging is a universal and progressive process involving the deterioration of physiological functions and the accumulation of cellular damage. Gene regulation programs influence how phenotypes respond to environmental and intrinsic changes during aging. Although several factors, including sex, are known to impact this process, the underlying mechanisms remain incompletely understood. Here, we investigate the functional organization patterns of skeletal muscle genes across different sexes and ages using gene co-expression networks (GCNs) to explore their influence on aging. We constructed GCNs for three different age groups for male and female samples, analyzed topological similarities and differences, inferred significant associated processes for each network, and constructed null models to provide statistically robust results. We found that each network is topologically and functionally distinct, with young women having the most associated processes, likely due to reproductive tasks. The functional organization and modularity of genes decline with age, starting from middle age, potentially leading to age-related deterioration. Women maintain better gene functional organization throughout life compared to men, especially in processes like macroautophagy and sarcomere organization. The study suggests that the loss of gene co-expression could be a universal aging marker. This research offers insights into how gene organization changes with age and sex, providing a complementary method to analyze aging.

ELSNC: A semi-supervised community detection method with integration of embedding-enhanced links and node content in attributed networks

In complex network analysis, detecting communities is becoming increasingly important. However, it is difficult to fuse multiple types of information to enhance the community-detection performance in real-world applications. Besides the nodes and the edges, a network also contains the structure of communities, its networking topological structure, and the network embeddings. Note that existing works on community detection have limited usage of all these information types in combination. In this work, we designed a novel unified model called embedding-enhanced link-based semi-supervised community detection with node content (ELSNC). ELSNC integrates the structure of the topology, the priori information, the network embeddings, and the node content. First, we employ two non-negative matrix factorization (NMF)–based stochastic models to characterize the node-community membership and the content-community membership (by performing similarity detection between a topic model and the NMF). Second, we introduce the nodes’ and networking embeddings’ topological similarity into the model as topological information. To model the topological similarity, we introduce a strong constraint (i.e., the priori information) and apply matrix completion to identify the community membership with the network embeddings’ representation ability. Finally, we present a semi-supervised community-detection method based on NMF that combines the network topology, content information, and the network embeddings. Our work’s innovation can be captured in two points: 1) As a type of semi-supervised community detection method, we extend the theory of semi-supervised methods on attributed networks and propose a unified model that integrates multiple information types. 2) The community membership obtained by the unified model simultaneously contains different information, including the topological, content, priori, and embedding information, which can more robustly be explored in the community structure in real-world scenarios. Furthermore, we performed a comprehensive evaluation of our proposed approach compared with state-of-the-art methods on both synthetic and real-world networks. The results show that our proposed method significantly outperformed the baseline methods.

Network Subgraph-based Method: Alignment-free Technique for Molecular Network Analysis

Background: Comparing directed networks using the alignment-free technique offers the advantage of detecting topologically similar regions that are independent of the network size or node identity. Objective: We propose a novel method to compare directed networks by decomposing the network into small modules, the so-called network subgraph approach, which is distinct from the network motif approach because it does not depend on null model assumptions. Methods: We developed an alignment-free algorithm called the Subgraph Identification Algorithm (SIA), which could generate all subgraphs that have five connected nodes (5-node subgraph). There were 9,364 such modules. Then, we applied the SIA method to examine 17 cancer networks and measured the similarity between the two networks by gauging the similarity level using Jensen- Shannon entropy (HJS). Results: We identified and examined the biological meaning of 5-node regulatory modules and pairs of cancer networks with the smallest HJS values. The two pairs of networks that show similar patterns are (i) endometrial cancer and hepatocellular carcinoma and (ii) breast cancer and pathways in cancer. Some studies have provided experimental data supporting the 5-node regulatory modules. Conclusion: Our method is an alignment-free approach that measures the topological similarity of 5-node regulatory modules and aligns two directed networks based on their topology. These modules capture complex interactions among multiple genes that cannot be detected using existing methods that only consider single-gene relations. We analyzed the biological relevance of the regulatory modules and used the subgraph method to identify the modules that shared the same topology across 2 cancer networks out of 17 cancer networks. We validated our findings using evidence from the literature.

DSC-Net: learning discriminative spatial contextual features for semantic segmentation of large-scale ancient architecture point clouds

Semantic segmentation of point cloud data of architectural cultural heritage is of significant importance for HBIM modeling, disease extraction and analysis, and heritage restoration research fields. In the semantic segmentation task of architectural point cloud data, especially for the protection and analysis of architectural cultural heritage, the previous deep learning methods have poor segmentation effects due to the complexity and unevenness of the data, the high geometric feature similarity between different components, and the large scale changes. To this end, this paper proposes a novel encoder-decoder architecture called DSC-Net. It consists of an encoder-decoder structure based on point random sampling and several fully connected layers for semantic segmentation. To overcome the loss of key features caused by random downsampling, DSC-Net has developed two new feature aggregation schemes: the enhanced dual attention pooling module and the global context feature module, to learn discriminative features for the challenging scenes mentioned above. The former fully considers the topology and semantic similarity of neighboring points, generating attention features that can distinguish categories with similar structures. The latter uses spatial location and neighboring volume ratio to provide an overall view of different types of architectural scenes, helping the network understand the spatial relationships and hierarchical structures between different architectural elements. The proposed modules can be easily embedded into various network architectures for point cloud semantic segmentation. We conducted experiments on multiple datasets, including the ancient architecture dataset, the ArCH architectural cultural heritage dataset, and the publicly available architectural segmentation dataset S3DIS. The results show that the mIoU reached 63.56%, 55.84%, and 71.03% respectively. The experimental results prove that our method has the best segmentation effect in dealing with challenging architectural cultural heritage data and also demonstrates its practicality in a wider range of architectural point cloud segmentation applications.

Angular Dependence of Hump‐Shape Hall Effects for Distinguishing between Karplus–Luttinger and Geometrical Origins

AbstractAmong the vast magnetic heterostructures explored in Condensed Matter Physics, two contrasting interpretations of the hump‐shaped Hall Effects remain ambiguous and debated, namely, the overlap of two opposite‐signed Karplus–Luttinger Hall loops associated with inhomogeneous collinear domains with perpendicular anisotropy, or the Geometrical/Topological Hall Effect emanated from hexagonal close‐packed lattice of Skyrmion ground state with smoothly varying non‐collinear moments. Their similarity in topology implies difficulty in discrimination via magnetic imaging. Here, this ambiguity is overcome and clarified by the divergence exponent of hump peak fields extracted from Hall measurements with magnetic field rotation on several heterostructures. Their difference in sensitivity to in‐plane fields reveals that the former mechanism involves higher uniaxial anisotropy than the latter, departing from the Skyrmion ground state regime by the Ginzburg–Landau framework of triple‐q spin‐wave superposition. Numerous material systems can be summarized into a single curve of divergence exponent versus the collinear quality factor, bridging the crossover of the two mentioned mechanisms.

Maritime Electro-Optical Image Object Matching Based on Improved YOLOv9

The offshore environment is complex during automatic target annotation at sea, and the difference between the focal lengths of visible and infrared sensors is large, thereby causing difficulties in matching multitarget electro-optical images at sea. This study proposes a target-matching method for visible and infrared images at sea based on decision-level topological relations. First, YOLOv9 is used to detect targets. To obtain markedly accurate target positions to establish accurate topological relations, the YOLOv9 model is improved for its poor accuracy for small targets, high computational complexity, and difficulty in deployment. To improve the detection accuracy of small targets, an additional small target detection head is added to detect shallow feature maps. From the perspective of reducing network size and achieving lightweight deployment, the Conv module in the model is replaced with DWConv, and the RepNCSPELAN4 module in the backbone network is replaced with the C3Ghost module. The replacements significantly reduce the number of parameters and computation volume of the model while retaining the feature extraction capability of the backbone network. Experimental results of the photovoltaic dataset show that the proposed method improves detection accuracy by 8%, while the computation and number of parameters of the model are reduced by 5.7% and 44.1%, respectively. Lastly, topological relationships are established for the target results, and targets in visible and infrared images are matched based on topological similarity.

Generating synthetic signaling networks for in silico modeling studies

Predictive models of signaling pathways have proven to be difficult to develop. Traditional approaches to developing mechanistic models rely on collecting experimental data and fitting a single model to that data. This approach works for simple systems but has proven unreliable for complex systems such as biological signaling networks. Thus, there is a need to develop new approaches to create predictive mechanistic models of complex systems. To meet this need, we developed a method for generating artificial signaling networks that were reasonably realistic and thus could be treated as ground truth models. These synthetic models could then be used to generate synthetic data for developing and testing algorithms designed to recover the underlying network topology and associated parameters. We defined the reaction degree and reaction distance to measure the topology of reaction networks, especially to consider enzymes. To determine whether our generated signaling networks displayed meaningful behavior, we compared them with signaling networks from the BioModels Database. This comparison indicated that our generated signaling networks had high topological similarities with BioModels signaling networks with respect to the reaction degree and distance distributions. In addition, our synthetic signaling networks had similar behavioral dynamics with respect to both steady states and oscillations, suggesting that our method generated synthetic signaling networks comparable with BioModels and thus could be useful for building network evaluation tools.

Attributed graph subspace clustering with residual compensation guided by adaptive dual manifold regularization

As an important technology for recommendation system, attributed graph clustering has received extensive attention recently. Attributed graph clustering methods based on graph neural networks are the mainstream methods . However, their assumption that the attributes of adjacent nodes are similar is often not satisfied in the real world, which influences the clustering performance. To this end, this paper proposes an attributed graph subspace clustering algorithm with residual compensation guided by adaptive dual manifold regularization (ADMRGC). On the basis of the low rank representation subspace clustering (LRR) model, ADMRGC introduces attributed manifold regularization, topological manifold regularization and residual compensation, which make ADMRGC possible to utilize attributed similarity and topological similarity at the same time, thus solving the problem that traditional subspace clustering only considers attributed information. In addition, ADMRGC balances the contribution of node attributes and topology by adaptively weighting the dual manifold regularization, and uses the residual representation matrix to describe the difference between node attributed similarity and topological neighbor relationship. Therefore, ADMRGC can avoid the limitation of graph neural network models assuming that the attributes of adjacent nodes are similar. Experimental results on 7 public graph datasets show that ADMRGC can achieve the best clustering performance on both high-homophily and low-homophily datasets. Especially, for datasets with low homophily, ADMRGC as a shallow model can also achieve better clustering performance than state-of-the-art deep neural network models.

SAILoR: Structure-Aware Inference of Logic Rules.

Boolean networks provide an effective mechanism for describing interactions and dynamics of gene regulatory networks (GRNs). Deriving accurate Boolean descriptions of GRNs is a challenging task. The number of experiments is usually much smaller than the number of genes. In addition, binarization leads to a loss of information and inconsistencies arise in binarized time-series data. The inference of Boolean networks from binarized time-series data alone often leads to complex and overfitted models. To obtain relevant Boolean models of gene regulatory networks, inference methods could incorporate data from multiple sources and prior knowledge in terms of general network structure and/or exact interactions. We propose the Boolean network inference method SAILoR (Structure-Aware Inference of Logic Rules). SAILoR incorporates time-series gene expression data in combination with provided reference networks to infer accurate Boolean models. SAILoR automatically extracts topological properties from reference networks. These can describe a more general structure of the GRN or can be more precise and describe specific interactions. SAILoR infers a Boolean network by learning from both continuous and binarized time-series data. It navigates between two main objectives, topological similarity to reference networks and correspondence with gene expression data. By incorporating the NSGA-II multi-objective genetic algorithm, SAILoR relies on the wisdom of crowds. Our results indicate that SAILoR can infer accurate and biologically relevant Boolean descriptions of GRNs from both a static and a dynamic perspective. We show that SAILoR improves the static accuracy of the inferred network compared to the network inference method dynGENIE3. Furthermore, we compared the performance of SAILoR with other Boolean network inference approaches including Best-Fit, REVEAL, MIBNI, GABNI, ATEN, and LogBTF. We have shown that by incorporating prior knowledge about the overall network structure, SAILoR can improve the structural correctness of the inferred Boolean networks while maintaining dynamic accuracy. To demonstrate the applicability of SAILoR, we inferred context-specific Boolean subnetworks of female Drosophila melanogaster before and after mating.

Peroxisome biogenesis factor PEX14 is crucial for survival and fecundity of female brown planthopper, Nilaparvata lugens (Stål)

Peroxisomes are ubiquitous cellular organelles participating in a variety of critical metabolic reactions. PEX14 is an essential peroxin responsible for peroxisome biogenesis. In this study, we identified the human PEX14 homolog in the brown planthopper, Nilaparvata lugens (Hemiptera: Delphacidae). N. lugens PEX14 (NlPEX14) showed significant topological similarity to its human counterpart. It is expressed throughout all developmental stages, with the highest expression observed in adult insects. Down-regulation of NlPEX14 through injection of NlPEX14-specific double-strand RNA impaired nymphal development. Moreover, females subjected to dsNlPEX14 treatment exhibited a significantly reduced lifespan. Additionally, we found abnormal ovarian development and a significant decrease in the number of eggs laid in NlPEX14-downregulated females. Further experiments support that the shortening of lifespan and the decrease in female fecundity can be attributed, at least partially, to the accumulation of fatty acids and reduced expression of vitellogenin. Together, our study reveals an indispensable function of NlPEX14 for insect reproduction and establishes a causal connection between the phenotypes and peroxisome biogenesis, shedding light on the importance of peroxisomes in female fecundity.

Integration of Semantic and Topological Structural Similarity Comparison for Entity Alignment without Pre-Training

Entity alignment (EA) is a critical task in integrating diverse knowledge graph (KG) data and plays a central role in data-driven AI applications. Traditional EA approaches rely on entity embeddings, but their effectiveness is limited by scarce KG input data and representation learning techniques. Large language models have shown promise, but face challenges such as high hardware requirements, large model sizes and computational inefficiency, which limit their applicability. To overcome these limitations, we propose an entity-alignment model that compares the similarity between entities by capturing both semantic and topological information to enable the alignment of entities with high similarity. First, we analyze descriptive information to quantify semantic similarity, including individual features such as types and attributes. Then, for topological analysis, we introduce four conditions based on graph connectivity and structural patterns to determine subgraph similarity within three hops of the entity’s neighborhood, thereby improving accuracy. Finally, we integrate semantic and topological similarity using a weighted approach that considers dataset features. Our model requires no pre-training and is designed to be compact and generalizable to different datasets. Experimental results on four standard EA datasets validate the effectiveness of our proposed model.

DMS: Low-overlap Registration of 3D Point Clouds with Double-layer Multi-scale Star-graph.

Registering 3D point clouds with low overlap is challenging in 3D computer vision, primarily due to difficulties in identifying small overlap regions and removing correspondence outliers. We observe that the neighborhood similarity can be utilized to detect point correspondence, and the consistent neighborhood correspondence can be used as a criterion to detect robust overlapping regions. So that a Double-layer Multi-scale Star-graph (DMS) structure is proposed to detect robust correspondences using two different types of multi-scale star-graphs. The first-layer Multi-scale Neighbor Feature Star-graphs (MNFS) takes each point as the center and its multi-scale nearest neighbors as the leaves. The MNFS enables to establish the initial correspondence candidate set between the two point clouds based on multi-scale neighborhood topology and feature similarity. Subsequently, each pair of corresponding points find their nearest neighbors within the correspondence sets to construct a Multi-scale Matching Star-graphs (MMS) on each side, so the mutual correspondence relationships between the MMS vertices are identified. These identified mutual correspondences are treated as vertices to construct the Multi-scale Correspondence Star-graphs (MCS), that indicate the relationships among the correspondences. We design edge weight and vertex weight criterion in MCS to detect only the robust correspondence set that has strong neighborhood consistency, so as to reject the outliers. Finally, the point cloud registration is conducted based on the detected robust correspondence. The experimental results demonstrate clearly that the proposed DMS method exhibits superior robustness when compared to existing state-of-the-art registration algorithms. The code of this study will be available at https://github.com/HualongCao/DMS.

On population-based structural health monitoring for bridges: Comparing similarity metrics and dynamic responses between sets of bridges

Bridges are valuable infrastructure assets that are challenging and expensive to maintain. State-of-the-art data-based bridge SHM solutions look to use bridge response data for condition assessment and damage detection. Data-based SHM methods can be limited in their application as they require large datasets to train models effectively, and most bridges lack the available data for the approaches to work. Further, it would be expensive and unrealistic to collect the required datasets to employ data-based methods to entire bridge networks. Recently, a population-based structural health monitoring (PBSHM) approach was proposed that seeks to leverage the data available for SHM problems by pooling together similar structures with their datasets. The PBSHM approach could be valuable in bridge SHM, enhancing the datasets available for ‘populations of bridges. The PBSHM approach for assessing the similarity of bridges has been considered before and was shown to be useful for identifying similar and different bridge types. However, no data were considered in the previous work, and similarity metrics were only qualified using engineering judgement. For the PBSHM approach to be useful in bridge SHM, there is still a need to check that ‘similar’ bridges have similar responses for transfer learning to be feasible. This paper expands upon previous work and provides originality by investigating if bridges identified as similar also exhibit similar responses. The PBSHM derived similarity metrics convey the topological similarity between structures, and mode shapes are identified as being a topologically sensitive bridge response. Therefore, a modal test campaign is carried out for a set of six real bridges, and Operational Modal Analysis is used to identify modal responses from each of the bridge decks. The Modal Assurance Criterion is used to evaluate the similarity between the mode shapes from pairs of bridges and is subsequently compared to the similarity metrics evaluated between those bridges. The similarity metrics were found to be reflective of the similarities identified between the respective bridges’ mode shapes for bridges of the same and different types. The significance of this finding is that it is an important step towards validating the PBSHM comparison approach for identify similar structures where transfer learning might be attempted.

Machine learning to identify structural motifs in asphaltenes

Asphaltenes are organic compounds that aggregate in crude oil with two dominant molecular architectures: archipelago and continental. Continental architectures possess a single uniform island structure composed of aromatic rings in contrast to archipelago architectures with aromatic cores interconnected through aliphatic chains. The structural composition of asphaltenes varies globally due to geographical differences, posing challenges in their classification due to a lack of uniformity. This study is the first known exploration of using image-based supervised machine learning, particularly the ResNet-50 neural network, for the binary classification of asphaltenes into continental and archipelago motifs. 255 continental and archipelago models underwent structural augmentations to create a sample size of 1,530 asphaltene structures that is robust enough for accurate results in both the training and testing portions of the machine learning. These augmentations included the repeated addition of carbons until a complete pentane chain was added to a specified carbon on each asphaltene structure. Using Mathematica, supervised ResNet-50 image-based classification was used on both original and augmented structure datasets to classify as either archipelago or continental. The classification was also implemented using topological similarity searching for association between atoms and the distance between them for further molecule identification. This study demonstrates the surprising effectiveness of image-based classification compared to traditional topological feature-based methods. Our results reveal that deep learning techniques, especially image-based approaches, provide novel and insightful ways to differentiate complex molecular structures like asphaltenes, challenging the traditional reliance on topological features alone. This research opens new avenues in chemical analysis and molecular characterization, highlighting the potential of machine learning in complex molecular systems.

The emergence and circulation of human immunodeficiency virus (HIV)-1 subtype C.

Introduction. Human immunodeficiency virus (HIV)-1 subtype C is the most prevalent globally and is thought to have originated in non-human primates in the Democratic Republic of Congo.Hypothesis/Gap Statement. Although the global dominance of HIV-1 subtype C is well established, a thorough understanding of its evolutionary history and transmission dynamics across various risk populations remains elusive. The current knowledge is insufficient to fully capture the global diversification and dissemination of this subtype.Aim. We for the first time sought to investigate the global evolutionary history and spatiotemporal dynamics of HIV-1 subtype C using a selection of maximum-likelihood-based phylodynamic approaches on a total of 1210 near full-length genomic sequences sampled from 32 countries, collected in 4 continents, with sampling dates between 1986-2019 among various risk groups were analysed.Methodology. We subsampled the HIV-1 subtype C genomic datasets based on continent and risk group traits, and performed nucleotide substitution model selection analysis, maximum likelihood (ML) phylogenetic reconstruction, phylogenetic tree topology similarity analysis, temporal signal analysis and traced the timings of viral spread both geographically and by risk group.Results. Based on the phylodynamic analyses of four datasets (full1210, locrisk626, loc562 and risk393), we inferred the time to the most recent common ancestor (TMRCA) in the 1930s and an evolutionary rate of 0.0023 substitutions per site per year. The total number of introduction events of HIV-1 subtype C between continents and between risk groups is estimated to be 71 and 115, respectively. The largest number of introductions occurred from Africa to Europe (n=32), from not-recorded to heterosexual (n=40) and from heterosexual to not-recorded (n=51) risk groups.Conclusion. Our results emphasize that HIV subtype C has mainly spread from Africa to Europe, likely through heterosexual transmission.

Manifold-based multi-graph embedding for semi-supervised classification

Graph semi-supervised learning (GSSL) plays an important role in semi-supervised classification by leveraging the similarity of graph topology and convex optimization with Laplacian-based regularization. Current approaches mainly focus on data representation in the Euclidean space and solely rely on a single graph structure, ignoring the underlined manifold-valued structure of the data. To this end, we propose a novel manifold-based multi-graph embedding (M2GE), which explicitly learns representations on both Riemannian manifold and Euclidean space in the embedding learning process. Incorporating multi-graph structure, the rich and varied features are used by M2GE to more fully capture categorical information, where the heterogeneous spatial representations from manifold are complementary. Extensive experimental results on three popular benchmarks verify demonstrate the superiority of our method over state-of-the art competitors.

Messages are Never Propagated Alone: Collaborative Hypergraph Neural Network for Time-Series Forecasting.

This paper delves into the problem of correlated time-series forecasting in practical applications, an area of growing interest in a multitude of fields such as stock price prediction and traffic demand analysis. Current methodologies primarily represent data using conventional graph structures, yet these fail to capture intricate structures with non-pairwise relationships. To address this challenge, we adopt dynamic hypergraphs in this study to better illustrate complex interactions, and introduce a novel hypergraph neural network model named CHNN for correlated time series forecasting. In more detail, CHNN leverages both semantic and topological similarities via an interaction model and hypergraph diffusion process, thereby constructing comprehensive collaborative correlation scores that effectively guide spatial message propagation. In addition, it incorporates short-term temporal information to generate efficient spatio-temporal feature maps. Lastly, a long-term temporal module is proposed to generate future predictions utilizing both temporal attention and a gated recurrent network. Comprehensive experiments conducted on four real-world datasets, i.e., Tiingo, Stocktwits, NYC-Taxi, and Social Network demonstrate that the proposed CHNN markedly outperforms a range of benchmark methods.