Graph Representation Research Articles

Advancing complex urban traffic forecasting: A fully attentional spatial-temporal network enhanced by graph representation

Accurate urban traffic forecasting is essential for intelligent transportation systems (ITS). However, the majority of existing forecasting methodologies predominantly concentrate on point-based forecasts (e.g., traffic detector forecasts). A limited number of them pay attention to the urban bidirectional road segments and the complex road network topology. To advance accurate traffic forecasting in complex urban scenarios, this paper proposes a Graph Representation enhanced Fully Attentional Spatial-Temporal network (GR-FAST). First, we construct a refined bidirectional road network graph (BRG) to depict the urban road network topology more accurately, particularly focusing on the turning patterns at intersections. Then, we adopt the graph representation methodology and introduce spatial information encoding (SIE) to explicitly characterize the significance of roads and network structure from multiple perspectives. Enhanced by SIE, spatial attention can capture spatial dependencies from both road network topologies and traffic pattern similarities, thereby forming a unified urban spatial cognition. Finally, a multi-scale residual perception (MRP) module is designed to balance the interplay of short-term temporal variability and long-term periodicity. Experiments on a real-world urban dataset from Wuhan, China, demonstrate that GR-FAST outperforms the state-of-the-art deep learning methods, achieving an improvement of 9.19%. Furthermore, ablation studies suggest that the explicit incorporation of complex road spatial topologies can significantly enhance forecasting accuracy.

Developing new methods for patient care pathway assessment for breast cancer patients in Hungary

Abstract Issue The idea of progressive healthcare models is to treat patients preferably in one location as close to their homes as possible. This contradicts the demands of efficiency: to centralise the availability of expensive procedures and highly specialised staff. Healthcare systems need to navigate between these challenges, especially in public health issues such as cancer care. Description of the problem Achieving the best possible health outcomes requires evidence-based policies that find balance between the demands of accessibility and efficiency. Therefore, we aimed to set up a methodology to review the healthcare model and assess patient movements across the system. We analysed the roaming paths of breast cancer patients diagnosed in Hungary in 2017. Inter-hospital roaming was represented as a graph: hospitals were the nodes, patient movements from one hospital to another were directed edges. This allowed us to apply network analysis: modularisation, weighted degrees calculation, etc. We then introduced the measure of terminality. A hospital is terminal in a given pathway if it is the last new one, i.e. all hospitals visited later had already been visited before. Terminality is defined as the proportion of terminal events among all pathways that cross a given hospital. The rationale of this is that the last hospital in cancer pathways is usually not the highest level one: after advanced treatments, patients are often referred back to local hospitals. Results Testing the method on Hungarian breast cancer patients we observed that results fit our prior expectations. The official ranking of hospitals on the progressivity scale (reflecting the availability of advanced treatments) shows sufficient agreement with terminality. Minor differences may point to situations where the healthcare system does not work fully according to regulations. Lessons Our methodology is recommended for system level patient pathway analysis in order to develop evidence-based healthcare policies. Key messages • Graph representation of inter-hospital patient roaming supports evidence-based policy developments for more efficient patient pathways in cancer care, regardless of the healthcare system design. • The applied new network analysis methodology can help assessing the real life processes of cancer care in a healthcare system through identifying the specifics of inter-hospital patient movements.

CAPTURE—Computational Analysis and Predictive Techniques for Urban Resource Efficiency

ABSTRACTMunicipal waste management (MWM) poses significant challenges in the context of rapid urbanisation and population growth. Accurate forecasting of waste production is crucial for designing sustainable waste management strategies. However, traditional forecasting methods often struggle to capture the complexities of waste generation dynamics. This paper proposes a novel methodology leveraging deep learning techniques to forecast municipal waste production. By harnessing the power of deep neural networks, our approach transcends the limitations of conventional models, providing more accurate and impactful predictions. We integrate heterogeneous data sources, including demographic and territorial information, into a comprehensive graph representation of municipalities. Graph Neural Networks are then employed to extract intricate spatial and temporal patterns from the graph structure. Empirical validation through a case study in the Apulia region demonstrates the effectiveness of our methodology in furnishing accurate forecasts for waste production. Our framework is adaptable and scalable, making it suitable for application across diverse geographical areas. This research contributes to advancing waste management practices by providing stakeholders with actionable insights for informed decision‐making.

DGCNN approach links metagenome-derived taxon and functional information providing insight into global soil organic carbon

Metagenomics can provide insight into the microbial taxa present in a sample and, through gene identification, the functional potential of the community. However, taxonomic and functional information are typically considered separately in downstream analyses. We develop interpretable machine learning (ML) approaches for modelling metagenomic data, combining the biological representation of species with their associated genetically encoded functions within models. We apply our methods to investigate soil organic carbon (SOC) stocks. First, we combine a diverse global set of soil microbiome samples with environmental data, improving the predictive performance of classic ML and providing new insights into the role of soil microbiomes in global carbon cycling. Our network analysis of predictive taxa identified by classical ML models provides context for their ecological significance, extending the focus beyond just the most predictive taxa to ‘hidden’ features within the model that might be considered less predictive using standard methods for explainability. We next develop unique graph representations for individual microbiomes, linking microbial taxa to their associated functions directly, enabling predictions of SOC via deep graph convolutional neural networks (DGCNNs). Interpretation of the DGCNNs distinguished between the importance of functions of key individual species, providing genome sequence differences, e.g., gene loss/acquisition, that associate with SOC. These approaches identify several members of the Verrucomicrobiaceae family and a range of genetically encoded functions, e.g., related to carbohydrate metabolism, as important for SOC stocks and effective global SOC predictors. These relatively understudied but widespread organisms could play an important role in SOC dynamics globally.

Diverse policy generation for the flexible job-shop scheduling problem via deep reinforcement learning with a novel graph representation

In scheduling problems common in the industry and various real-world scenarios, responding in real-time to disruptive events is important. Recent methods propose the use of deep reinforcement learning (DRL) to learn policies capable of generating solutions under this constraint. However, current DRL approaches struggle with large instances, which are common in real-world scenarios. The objective of this paper is to introduce a new DRL method for solving the flexible job-shop scheduling problem, with a focus on these type of instances. The approach is based on the use of heterogeneous graph neural networks to a more informative graph representation of the problem. This novel modeling of the problem enhances the policy’s ability to capture state information and improve its decision-making capacity. Additionally, we introduce two novel approaches to enhance the performance of the DRL approach: the first involves generating a diverse set of scheduling policies, while the second combines DRL with dispatching rules (DRs) constraining the action space, with a variable degree of freedom depending on the chosen policy. Experimental results on two public benchmarks show that our approach outperforms DRs and achieves superior results compared to three state-of-the-art DRL methods, particularly for large instances.

RNA fold prediction by Monte Carlo in graph space and the statistical mechanics of tertiary interactions.

Using a graph representation of RNA structures, we have studied the ensembles of secondary and tertiary graphs two sets of RNA with Monte Carlo simulations. The first consisted of 91 target ribozyme and riboswitch sequences of moderate lengths (< 150 nt) having a variety of secondary, H-type pseudoknots and kissing loop interactions. The second set consisted of 71 more diverse sequences across many RNA families. Using a simple empirical energy model for tertiary interactions and only sequence information for each target as input, the simulations examined how tertiary interactions impact the statistical mechanics of the fold ensembles. The results show that the graphs proliferate enormously when tertiary interactions are possible, producing an entropic driving force for the ensemble to access folds having tertiary structures even though they are overall energetically unfavorable in the energy model. For each of the targets in the two test sets, we assessed the quality of the model and the simulations by examining how well the simulated structures were able to predict the native fold and compared the results to fold predictions from ViennaRNA. Our model generated good or excellent predictions in a large majority of the targets. Overall, this method was able to produce predictions of comparable quality to Vienna, but it outperformed Vienna for structures with H-type pseudoknots. The results suggest that while tertiary interactions are predicated on real-space contacts, their impacts on the folded structure of RNA can be captured by graph space information for sequences of moderate lengths, using a simple tertiary energy model for the loops, the base pairs and base stacks.

Spatial domains identification in spatial transcriptomics using modality-aware and subspace-enhanced graph contrastive learning

Spatial transcriptomics (ST) technologies have emerged as an effective tool to identify the spatial architecture of tissues, facilitating a comprehensive understanding of organ function and the tissue microenvironment. Spatial domain identification is the first and most critical step in ST data analysis, which requires thoughtful utilization of tissue microenvironment and morphological priors. Here, we propose a graph contrastive learning framework, GRAS4T, which combines contrastive learning and a subspace analysis model to accurately distinguish different spatial domains by capturing the tissue microenvironment through self-expressiveness of spots within the same domain. To uncover the pertinent features for spatial domain identification, GRAS4T employs a graph augmentation based on histological image priors, preserving structural information crucial for the clustering task. Experimental results on eight ST datasets from five different platforms show that GRAS4T outperforms five state-of-the-art competing methods. Significantly, GRAS4T excels at separating distinct tissue structures and unveiling more detailed spatial domains. GRAS4T combines the advantages of subspace analysis and graph representation learning with extensibility, making it an ideal framework for ST domain identification.

Encoder embedding for general graph and node classification

Graph encoder embedding, a recent technique for graph data, offers speed and scalability in producing vertex-level representations from binary graphs. In this paper, we extend the applicability of this method to a general graph model, which includes weighted graphs, distance matrices, and kernel matrices. We prove that the encoder embedding satisfies the law of large numbers and the central limit theorem on a per-observation basis. Under certain condition, it achieves asymptotic normality on a per-class basis, enabling optimal classification through discriminant analysis. These theoretical findings are validated through a series of experiments involving weighted graphs, as well as text and image data transformed into general graph representations using appropriate distance metrics.

Multimodal Representation Learning via Graph Isomorphism Network for Toxicity Multitask Learning.

Toxicity is paramount for comprehending compound properties, particularly in the early stages of drug design. Due to the diversity and complexity of toxic effects, it became a challenge to compute compound toxicity tasks. To address this issue, we propose a multimodal representation learning model, termed multimodal graph isomorphism network (MMGIN), to address this challenge for compound toxicity multitask learning. Based on fingerprints and molecular graphs of compounds, our MMGIN model incorporates a multimodal representation learning model to acquire a comprehensive compound representation. This model adopts a two-channel structure to independently learn fingerprint representation and molecular graph representation. Subsequently, two feedforward neural networks utilize the learned multimodal compound representation to perform multitask learning, encompassing compound toxicity classification and multiple compound category classification simultaneously. To test the effectiveness of our model, we constructed a novel data set, termed the compound toxicity multitask learning (CTMTL) data set, derived from the TOXRIC data set. We compare our MMGIN model with other representative machine learning and deep learning models on the CTMTL and Tox21 data sets. The experimental results demonstrate significant advancements achieved by our MMGIN model. Furthermore, the ablation study underscores the effectiveness of the introduced fingerprints, molecular graphs, the multimodal representation learning model, and the multitask learning model, showcasing the model's superior predictive capability and robustness.

Early detection of dementia through retinal imaging and trustworthy AI

Alzheimer’s disease (AD) is a global healthcare challenge lacking a simple and affordable detection method. We propose a novel deep learning framework, Eye-AD, to detect Early-onset Alzheimer’s Disease (EOAD) and Mild Cognitive Impairment (MCI) using OCTA images of retinal microvasculature and choriocapillaris. Eye-AD employs a multilevel graph representation to analyze intra- and inter-instance relationships in retinal layers. Using 5751 OCTA images from 1671 participants in a multi-center study, our model demonstrated superior performance in EOAD (internal data: AUC = 0.9355, external data: AUC = 0.9007) and MCI detection (internal data: AUC = 0.8630, external data: AUC = 0.8037). Furthermore, we explored the associations between retinal structural biomarkers in OCTA images and EOAD/MCI, and the results align well with the conclusions drawn from our deep learning interpretability analysis. Our findings provide further evidence that retinal OCTA imaging, coupled with artificial intelligence, will serve as a rapid, noninvasive, and affordable dementia detection.

Multi-particle quantum walks on 3D integrated photonic chip

Quantum walks provide a speed-up in computational power for various quantum algorithms and serve as inspiration for the construction of complex graph representations. Many pioneering works have been dedicated to expanding the experimental state space and the complexity of graphs. However, these experiments are mostly limited to small experimental scale, which do not reach a many-body level and fail to reflect the multi-particle quantum interference effects among non-adjacent modes. Here, we present a quantum walk with three photons on a two-dimensional triangular lattice, which is mapped to a 19 × 19 × 19 high-dimensional state space and constructs a complex graph with 6859 nodes and 45,486 edges. By utilizing the statistical signatures of the output combinations and incorporating machine learning techniques, we successfully validate the nonclassical properties of the experiment. Our implementation provides a paradigm for exponentially expanding the state space and graph complexity of quantum walks, paving the way for surmounting the classical regime in large-scale quantum simulations.

Layered Graph Models for the Electric Vehicle Routing Problem With Nonlinear Charging Functions

AbstractElectric vehicle routing problems (EVRPs) involve the routing of a fleet of electric vehicles (EVs) to visit a set of customers while typically minimizing the total travel and charging time. Due to their limited autonomy, EVs may need to recharge their batteries en‐route at charging stations (CSs). Thus, routing decisions also include which CSs to visit, and how much energy to charge during those visits. These decisions are compounded by the fact that charging times follow a nonlinear charging function with respect to the EV's state of charge (SoC). We propose a layered graph representation for the EVRP with nonlinear charging functions (EVRP‐NL). Specifically, the layers correspond to discretized SoC values. Therefore, the arcs' energy consumption is approximated to match those values. We develop two compact formulations based on the layered graph. Furthermore, we introduced two charging policies that facilitate aligning charging duration with practical considerations. Computational results demonstrate the effectiveness of our formulations. Our best formulation effectively handles instances with up to 40 customers. On those instances, compared to the state‐of‐the‐art compact formulation, our formulation solves 13 more instances to optimality with less than half of the computational time. Considering instances solved by both formulations to optimally, the approximation entailed by our formulation yields a 0.94% deviation on average. Since our best performing formulation is compact, it may be readily used by a broad audience. Furthermore, as the majority of algorithms for the EVPR and its variants are heuristics, our formulation could be beneficial in evaluating the performance of these methods.

Online coloring of disk graphs

In this paper, we give a family of online algorithms for the classical coloring problem of intersection graphs of disks with bounded diameter. Our algorithms make use of a geometric representation of such graphs and are inspired by an algorithm of Fiala et al., but have better competitive ratios. The improvement comes from using two techniques of partitioning the set of vertices before coloring them. One of them is an application of a b-fold coloring of the plane. The method is more general and we show how it can be applied to coloring other shapes on the plane as well as adjust it for online L(2,1)-labeling.

Tabular affiliation extraction based on graph convolutional network

This paper studies the problem of extracting affiliation relationships between cells in a table in the field of table recognition and analysis. The task of extracting affiliation relationships between tables is defined. Combining the similarity between table and graph structure, a graph representation method for cells in a table is given, and a graph convolutional network-based affiliation relationshipextraction model is proposed. The model aggregates features of cells in a table and their neighboring cells through a graph convolutional network, predicts whether there is an affiliation relationship between cells, and realizes relationship extraction. In order to verify the effectiveness of the model, two datasets, Rel-forms for Chinese forms and Rel-SciTSR for English forms, are annotated. Through experiments, the F1 scores on the above two datasets and the joint dataset reached , respectively 98.61%、96.55%、97.05%, verifying the effectiveness of the affiliation relationship extraction model on these two datasets, and analyzing the influence of different factors such as text content, coordinate information, cell attributes, and relative direction between cells on the experimental results of affiliation relationship extraction.

Adversarial nonnegative matrix factorization for temporal link prediction

Temporal link prediction has been extensively studied and widely applied in various applications, aiming to predict future network links based on the historical networks. However, most existing methods ignore the behavior of previous network updating information in temporal networks. To address these issues, we propose a novel link prediction model based on adversarial nonnegative matrix factorization, which fuses graph representation and adversarial learning to perform temporal link prediction. Specifically, we add a bounded adversary matrix to the input matrix to provide the robustness against real perturbations. Then, our model fully exploits the impact of snapshots by using communicability. Simultaneously, we utilize the cosine similarity to extract the node similarity and map it to low-dimensional latent representation to preserve the local structure. Additionally, we provide effective updating rules to learn the parameters of this model. Extensive experiments results on six real-world networks demonstrate that the proposed method outperforms several classical and the state-of-art matrix-based methods.

Compositionally restricted atomistic line graph neural network for improved thermoelectric transport property predictions

Graph neural networks (GNNs) have evolved many variants for predicting the properties of crystal materials. While most networks within this family focus on improving model structures, the significance of atomistic features has not received adequate attention. In this study, we constructed an atomistic line GNN model using compositionally restricted atomistic representations which are more elaborate set of descriptors compared to previous GNN models, and employing unit graph representations that account for all symmetries. The developed model, named as CraLiGNN, outperforms previous representative GNN models in predicting the Seebeck coefficient, electrical conductivity, and electronic thermal conductivity that are recorded in a widely used thermoelectric properties database, confirming the importance of atomistic representations. The CraLiGNN model allows optional inclusion of additional features. The supplement of bandgap significantly enhances the model performance, for example, more than 35% reduction of mean absolute error in the case of 600 K and 1019 cm−3 concentration. We applied CraLiGNN to predict the unrecorded thermoelectric transport properties of 14 half-Heusler and 52 perovskite compounds, and compared the results with first-principles calculations, showing that the model has extrapolation ability to identify the thermoelectric potential of materials.

ERL-ProLiGraph: Enhanced representation learning on protein-ligand graph structured data for binding affinity prediction.

Predicting Protein-Ligand Binding Affinity (PLBA) is pivotal in drug development, as accurate estimations of PLBA expedite the identification of promising drug candidates for specific targets, thereby accelerating the drug discovery process. Despite substantial advancements in PLBA prediction, developing an efficient and more accurate method remains non-trivial. Unlike previous computer-aid PLBA studies which primarily using ligand SMILES and protein sequences represented as strings, this research introduces a Deep Learning-based method, the Enhanced Representation Learning on Protein-Ligand Graph Structured data for Binding Affinity Prediction (ERL-ProLiGraph). The unique aspect of this method is the use of graph representations for both proteins and ligands, intending to learn structural information continued from both to enhance the accuracy of PLBA predictions. In these graphs, nodes represent atomic structures, while edges depict chemical bonds and spatial relationship. The proposed model, leveraging deep-learning algorithms, effectively learns to correlate these graphical representations with binding affinities. This graph-based representations approach enhances the model's ability to capture the complex molecular interactions critical in PLBA. This work represents a promising advancement in computational techniques for protein-ligand binding prediction, offering a potential path toward more efficient and accurate predictions in drug development. Comparative analysis indicates that the proposed ERL-ProLiGraph outperforms previous models, showcasing notable efficacy and providing a more suitable approach for accurate PLBA predictions.

Telomere-to-telomere assembly by preserving contained reads.

Automated telomere-to-telomere (T2T) de novo assembly of diploid and polyploid genomes remains a formidable task. A string graph is a commonly used assembly graph representation in the assembly algorithms. The string graph formulation employs graph simplification heuristics, which drastically reduce the count of vertices and edges. One of these heuristics involves removing the reads contained in longer reads. In practice, this heuristic occasionally introduces gaps in the assembly by removing all reads that cover one or more genome intervals. The factors contributing to such gaps remain poorly understood. In this work, we mathematically derived the frequency of observing a gap near a germline and a somatic heterozygous variant locus. Our analysis shows that (i) an assembly gap due to contained read deletion is an order of magnitude more frequent in Oxford Nanopore reads than PacBio HiFi reads due to differences in their read-length distributions, and (ii) this frequency decreases with an increase in the sequencing depth. Drawing cues from these observations, we addressed the weakness of the string graph formulation by developing the RAFT assembly algorithm. RAFT addresses the issue of contained reads by fragmenting reads and producing a more uniform read-length distribution. The algorithm retains spanned repeats in the reads during the fragmentation. We empirically demonstrate that RAFT significantly reduces the number of gaps using simulated datasets. Using real Oxford Nanopore and PacBio HiFi datasets of the HG002 human genome, we achieved a twofold increase in the contig NG50 and the number of haplotype-resolved T2T contigs compared to Hifiasm.

Maximizing wind farm production through pitch control using graph neural networks and hybrid learning methods

AbstractThis article presents a novel methodology to maximize wind farm power generation by integrating graph neural networks (GNN), supervised learning, and reinforcement learning techniques. First, the article introduces a graph‐based representation of the wind farm, capturing wind turbines as vertices and the inter‐turbine wake interactions as edges. The construction of this graph representation integrates the Jensen wake model, which includes insights derived from prior knowledge of wind farm aerodynamics. Subsequently, a detailed description of the GNN model's architecture, incorporating a message passing mechanism, is outlined. This GNN model is trained initially with supervised learning using a dataset of optimal pitch angles generated from the analytical results derived from Jensen wake model. Moreover, to improve the GNN model's accuracy and adaptability, reinforcement learning techniques are employed. The GNN model interacts with a high‐fidelity wind farm simulation environment, receiving feedback in the form of rewards derived from the wind farm's actual power output. Through a policy gradient approach, the GNN parameters undergo iterative updates, enabling the model to learn and adapt to dynamic wind conditions and intricate turbine interactions. The effectiveness and advantages of the proposed methodology are demonstrated through comprehensive case studies across various wind farm layouts.

GeoHi-GNN: Geometry-aware hierarchical graph representation learning for normal estimation

Normal estimation has been one of the key tasks in point cloud analysis, while it is challenging when facing with severe noises or complex regions. The challenges mainly come from the selection of supporting points for estimation, that is, improper selections of points and points' scale will lead to insufficient information, loss of details, etc. To this end, this paper proposes one feature-centric fitting scheme, GeoHi-GNN, by learning geometry-aware hierarchical graph representation for fitting weights estimation. The main functional module is the continuously conducted Hierarchically Geometric-aware (HG) module, consisting of two core operations, namely, the graph node construction (GNC) and the geometric-aware dynamic graph convolution (GDGC). GNC aims to aggregate the feature information onto a smaller number of nodes, providing global-to-local information while avoiding the interferences from noises in larger scales. With these nodes distributed in different scales, GDGC dynamically updates the node features regarding to both intrinsic feature and extrinsic geometric information. Finally, the hierarchical graphical features are cascaded to estimate the weights for supporting points in the surface fitting. Through the extensive experiments and comprehensive comparisons with the state-of-the-arts, our scheme has exhibited many attractive advantages such as being geometry-aware and robust, empowering further applications like more accurate surface reconstruction.