Conventional data visualization techniques in single-cell analysis (such as two-dimensional dot plots, SPADE, PCA, t-SNE, or UMAP) often fall short in enabling an intuitive understanding of high-parameter flow cytometry data. These methods tend to oversimplify complex biological relationships, lack biologically meaningful interpretations, and offer no principled framework for downstream quantitative analysis. To address these limitations, we present a graph-based (network-based) visualization framework grounded in optimal transport theory. In this framework, cell populations are defined by their marker-expression profiles, and inter-population similarity is quantified using an efficiently computable optimal transport formulation known as the Sinkhorn distance. Our approach produces biologically consistent two-dimensional graph layouts using a phenotype-aware Hamming distance. Structural differences between sample graphs are characterized through a customized graph-edit distance that captures changes in population size, marker expression, and relationships between populations. We demonstrate our methods on two flow cytometry datasets: one from a clinical trial of dendritic cell-based immunotherapy in malignant peritoneal mesothelioma, involving 14 patients sampled at three time points with 14-color panels, and another from FlowCAP-II, which involved 43 acute myeloid leukemia patient samples analyzed with 7-color panels. Our framework produces robust, quantitative visual summaries of cell populations and supports statistical analysis based on graph edit distances, thereby offering new insights into disease progression and treatment response. Ultimately, our method bridges the gap between flow cytometry data visualization and biological interpretation.

ABSTRACT With the rapid growth of high-resolution remote sensing data, efficient and accurate image retrieval has become increasingly important for urban analysis and land management. Traditional retrieval approaches based on global or local features often suffer from low efficiency and limited robustness when applied to large-scale, multi-source datasets. To address these challenges, we propose a multi-scale image retrieval framework that integrates optimized building contour analysis with graph-based similarity measurement. Building footprints are first extracted using hybrid CNN – Transformer segmentation and then refined through a series of contour optimization steps, including adaptive simplification, rectification, and structural reconstruction. The optimized polygons are encoded by shape descriptors for preliminary screening, while a Siamese network further evaluates polygon similarity. At the final stage, building distribution is modelled as a labelled undirected graph, and an improved graph edit distance algorithm is applied for accurate matching. The framework was evaluated on Google, Bing, and ArcGIS satellite imagery as well as several public datasets. Experimental results demonstrate that the proposed method substantially improves retrieval accuracy and computational efficiency, while preserving geometric fidelity of complex urban structures. This study highlights the potential of contour- and graph-based approaches for advancing remote sensing image retrieval, with promising applications in urban planning, environmental monitoring, and disaster assessment.

Student dropout remains a persistent issue in higher education, affecting institutional effectiveness and student success rates. This paper proposes a graph-based predictive model that employs subgraph matching and Graph Edit Distance (GED) to identify students at high risk of dropout. By modeling students and courses as an undirected bipartite graph, the system detects structural similarities between student profiles. The proposed model was evaluated using a dataset of 282 students from a private university in Indonesia under three data-splitting scenarios: 70/30, 79/21, and 89/11. Evaluation metrics include precision, recall, F1-score, and accuracy. The model achieved its best performance at the 89/11 split, with an accuracy of 91%, precision of 1.00, recall of 0.88, and an F1-score of 0.94. Results suggest that increasing the proportion of training data enhances generalization and prediction accuracy. GED demonstrated effectiveness in capturing subtle structural distinctions among student–course relationships, enabling early dropout risk identification. The primary contribution of this study is the development of a graph-analytic framework for dropout prediction, offering an alternative to traditional models such as logistic regression and decision trees that lack relational awareness. Future work will incorporate behavioral and socio-economic attributes to further improve prediction outcomes.

To address the challenge of securely computing graph edit distance in directed graph scenarios without compromising parties' privacy data, this paper proposes a homomorphic encryption-based, dual-cloud-assisted secure graph edit distance computation protocol. The protocol encodes graph structures as adjacency matrices and introduces random parity noise to protect the original data's parity. It then leverages homomorphic encryption to compute edit distance in ciphertext form. Employing a dual-cloud server architecture, the protocol offloads most computational tasks to the cloud, eliminating direct interaction between data users. This approach enhances computational efficiency while ensuring privacy security. Theoretical analysis and simulation experiments demonstrate the protocol's security under a semi-honest model, with superior user-side computational overhead compared to existing solutions.

Measuring similarity between complex objects is a fundamental task in many scientific fields. When objects are represented as graphs, graph similarity/distance measures offer a powerful framework for quantifying structural resemblance. Those comparative measures play a key role in domains such as network science, chemoinformatics, and social network analysis. While methods like graph edit distance and graph kernels are widely used, they can be computationally intensive or fail to capture fine structural variations, since they require graphs without any structural uncertainty. Another class of methods is based on using topological indices to encode structural information of the graphs, followed by the application of distance or similarity measures for real numbers to obtain corresponding graph-level metrics. In this paper, we introduce a novel class of distance/similarity measures which are based on multiple topological indices. In many cases, they can be computed in polynomial time, which makes our method computationally efficient in practice. We demonstrate its effectiveness through comparisons and show that it captures subtle structural information meaningfully. Additionally, we explore its applicability in two domains: analyzing random graph models in network theory and assessing molecular similarity among isomers in chemoinformatics. These preliminary results suggest that our approach holds promise for graph comparison across disciplines.

Metadata extraction from natural history collection labels is a pivotal task for the online publication of digitized specimens. However, given the scale of these collections-which are estimated to host more than 2 billion specimens worldwide, including ca. 400 million herbarium specimens-manual metadata extraction is an extremely time-consuming task. Thus, automated data extraction from digital images of specimens and their labels therefore is a promising application of state-of-the-art computer vision techniques. Extracting information from herbarium specimen labels normally involves three main steps: text segmentation, multilingual and handwriting recognition, and data parsing. The primary bottleneck in this workflow lies in the limitations of Optical Character Recognition (OCR) systems. This study explores how the general knowledge embedded in multimodal Transformer models can be transferred to the specific task of herbarium specimen label digitization. The final goal is to develop an easy-to-use, end-to-end solution to mitigate the limitations of classic OCR approaches while offering greater flexibility to adapt to different label formats. Donut-base, a pre-trained visual document understanding (VDU) transformer, was the base model selected for fine-tuning. A dataset from the University of Pisa served as a test bed. The initial attempt achieved an accuracy of 85%, measured using the Tree Edit Distance (TED), demonstrating the feasibility of fine-tuning for this task. Cases with low accuracies were also investigated to identify limitations of the approach. In particular, specimens with multiple labels, especially if combining handwritten and typewritten text, proved to be the most challenging. Strategies aimed at addressing these weaknesses are discussed.

Decision tree models are widely applied in educational data analysis due to their simplicity and interpretability. However, these models exhibit high sensitivity to variations in training data, where minor sampling changes can result in substantially different tree structures, potentially reducing model reliability and consistency. This study aims to empirically investigate the structural robustness of decision trees under variations in educational data sampling and to evaluate strategies for improving model stability. An experimental framework is implemented using several educational datasets processed through random subsampling, bootstrap resampling, and k-fold cross-validation. Structural robustness is quantitatively assessed using tree edit distance, node similarity ratio, and tree depth variability. The results indicate that small sampling perturbations can cause significant structural divergence, particularly in datasets characterized by high noise levels and feature correlations. Nevertheless, pruning techniques and ensemble-based stabilization methods effectively enhance structural consistency and reduce model variance. These findings highlight the importance of robustness evaluation in educational data mining and provide empirical insights for developing reliable, stable, and interpretable decision-support systems in educational environments.

Current methods that generate a scene graph of a given image create an overdense graph regarding the number of objects and relationships found in the image, making the graph less effective in describing what is relevant in the image and, consequently, in downstream tasks such as cross-modal retrieval and mining. In this work, we propose a novel method that generates a scene graph that reflects what humans find important when describing an image. During training, our scene generation method is guided by human-drafted captions describing the images, which we assume will focus on essential scene elements. This guidance is realized by properly designed loss functions. During inference, scene graphs are generated solely from images. We evaluate the resulting scene graphs by comparing them with the ground-truth scene graphs of Visual Genome that are created by humans. Evaluation is done with recall- and precision-oriented metrics and graph edit distances. In the first set of experiments, we benchmark existing scene graph generation models, then we add the newly proposed loss functions leading to improved performance, especially in terms of the graph edit distance. Extra experiments show that the correct recognition of unimportant background objects and their relationships is crucial when generating human-like scene graphs. The codebase is released on github:https://github.com/VSJMilewski/relevance_graphs

Molecular generation is a critical method in drug design, but its practical application is often limited by the difficulty of synthesizing the generated molecules. To address this challenge, we present RetroScore, a synthetic accessibility evaluation framework guided by multistep retrosynthesis. Our methodology integrates the semi-template model Graph2Edits with the multistep retrosynthesis planning algorithm Retro*, forming the Graph2Edits-Retro*d system. By incorporating the green chemistry metric of graph edit distance into the reaction cost function and a multistage screening protocol, this system identifies optimal routes while balancing reliability, synthetic efficiency, and economic feasibility. Benchmark evaluations demonstrate a 97.37% planning success rate with balanced optimization across route length, confidence score, and graph edit distance. In the molecular generation task, the RetroScore outperforms six of the seven synthetic accessibility metrics, yielding molecules with enhanced synthetic accessibility profiles across heterogeneous evaluation frameworks. To facilitate practical implementation, we developed an open-access web platform for automated retrosynthesis route prediction and RetroScore calculation, providing researchers with rapid synthetic accessibility assessments. The RetroScore web server is publicly accessible at http://aidd.bioai-global.com/RetroScore/, and the source code is available at https://github.com/Snowgao320/RetroScore.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13321-025-01138-6.

Graph similarity computation is one of the fundamental challenges in graph data mining, with critical applications in chemical molecule analysis and social network analysis. Traditional methods, such as graph edit distance and maximum common subgraph, are NP-hard and computationally expensive. Although neural network-based approaches have improved efficiency and accuracy, existing methods still exhibit limitations: they often focus excessively on global features while neglecting local substructures, or vice versa, and fail to evaluate the relative importance of node-level, subgraph level, and global features. To address these issues, this paper proposes Similarity Subgraph Global Tensor Network(SSGTN), a novel graph similarity computation framework that integrates subgraph and global features. Specifically, SSGTN employs graph convolutional networks to extract subgraph features and graph attention mechanisms to capture global representations. A Subgraph Global Tensor Network is then designed to dynamically fuse pooled subgraph features with global features, thereby jointly optimizing local and global characterizations. Finally, all features are aggregated to generate the graph similarity score. Experiments on real-world datasets demonstrate that SSGTN outperforms baseline models across three evaluation metrics, confirming its efficiency and accuracy.

Graph edit distance (GED) is an important metric for measuring the distance or similarity between two graphs. It is defined as the minimum number of edit operations required to transform one graph into another. Computing the exact GED between two graphs is an NP-hard problem. With the success of deep learning across various application domains, graph neural networks have also been recently utilized to predict the GED between graphs. However, the existing studies on learning-based methods have two significant limitations. (1)~The development of deep learning models for GED prediction has been explored in various research fields (e.g., databases, machine learning, information retrieval, and computer vision), yet cross-field evaluations have been quite limited. (2)~More importantly, all these advancements have been evaluated against a simple combinatorial heuristic baseline, with their models shown to outperform it. In this paper, we aim to bridge this knowledge gap. We first conduct a holistic review of the existing learning-based methods, categorizing them into non-interpretable and interpretable GED prediction approaches, while highlighting their overarching design principles and relationships among these models. Secondly, we present a simple yet effective combinatorial heuristic algorithm App-BMao for GED estimation, adapted from an existing exact GED computation algorithm. App-BMao provides interpretable GED estimation with controlled time and space complexity. Extensive empirical evaluations on three widely used datasets show that the new heuristic algorithm App-BMao outperforms all existing learning-based approaches for both interpretable and non-interpretable GED prediction.

Table structure recognition, as the most important subtask in image-based table recognition, recognizes the logical relationship between adjacent cells and represents the structured essence of the table. However, in reality, tables often appear in many different styles, such as those containing many blank cells, large row–column spans, missing separator lines, or even no separator lines at all. Different styles present great challenges to table structure recognition. In response to the above situation, we carefully analyzed the structural characteristics of the table and believed that rows and columns are their inherent characteristics. However, row–column annotations are rarely encountered in existing publicly available datasets, and cells with row–column spans are often ignored. Thus, we propose an innovative table structure annotation scheme, where the objects to be annotated include rows, columns, and cells with row–column spans. Furthermore, we released a challenging dataset named Row–Column Segmentation for Table Structure Recognition (RCSTSR), which contains more than 12,000 table images of different styles, each of which is notated with the corresponding mask. Then, on the basis of this dataset, we construct an effective semantic segmentation-based solution for table structure recognition. It consists of two main parts: an improved masked-attention mask transformer model, named TableStructureFormer, and corresponding postprocessing. Among them, the former is responsible for predicting the masks of objects in the table image, and the latter is used to generate the table structure on the basis of the predicted masks. Considering that long-distance feature maps are more useful than local feature maps for row–column segmentation, we propose the dual-path adaptive weighted attention module to aggregate the multilevel long-distance feature maps and adaptively select more input information by introducing enhanced strip pooling and learnable weighted parameters, thereby improving the segmentation performance. In addition, to address the difficulty of perfectly segmenting the details of rows and columns, we propose the deep detail supervision module to guide the segmentation model to learn the detailed feature maps about objects, thereby further correcting their masks. The experimental results show that for row–column segmentation, the mean intersection over union (mIoU) values of TableStructureFormer are 92.15% and 92.46%, respectively, which are significantly greater than those of the existing segmentation models. For table structure recognition, it can also accurately generate table structures according to the segmentation results. The tree edit distance similarity for the table structure (TEDS-Struct), precision, and recall are 91.42%, 90.96%, and 87.73%, respectively, indicating excellent table logic understanding ability. When the position matching thresholds are 0.50 and 0.75, the average accuracy (AP) is 91.50% and 86.32%, respectively, indicating excellent spatial perception ability. In summary, the proposed TableStructureFormer can accurately and robustly implement row–column segmentation and table structure recognition, even when facing different styles of tables.

Graph Edit Distance (GED) is a widely recognized metric for measuring graph similarity, yet its NP-complete nature poses challenges for fast and accurate computation. This paper introduces FGWAlign, an Optimal Transport (OT)-based approach for graph alignment and GED computation. We take the first step to theoretically demonstrate and that computing GED can be transformed into optimizing a particular OT variant—the Fused Gromov-Wasserstein distance. Tailored to the GED problem structure, we further implement three key enhancements to the standard FGW solver: (1) a random exploration scheme to better locate the global optimum, (2) a diverse projection strategy for post-processing the transportation plan to escape local optima, and (3) a novel extension to accommodate multi-relational graphs with edge labels. With O (| V || E |) time complexity and O (| V | 2 ) space complexity, where | V | and | E | are the maximum number of nodes and edges between the two compared graphs, FGWAlign achieves a superior balance of efficiency, accuracy, and scalability. Empirical results show that, compared with 12 representative GED computation methods across different categories on 4 real-world graph datasets, FGWAlign reduces computation errors by over 80% and achieves 15–60× speedup. It also demonstrates promising resutls on downstream applications including labeled graph alignment and graph-level anomaly detection, highlighting its versatility. FGWAlign opens up promising avenues for future applications in graph data management.

MotivationPangenome variation graphs are an increasingly used tool to perform genome analysis, aiming to replace a linear reference in a wide variety of genomic analyses. The construction of a variation graph from a collection of chromosome-size genome sequences is a difficult task that is generally addressed using a number of heuristics. The question that arises is to what extent the construction method influences the resulting graph, and the characterization of variability.ResultsWe aim to characterize the differences between variation graphs derived from the same set of genomes with a metric which expresses and pinpoint differences. We designed a pairwise variation graph comparison algorithm, which establishes an edit distance between variation graphs, threading the genomes through both graphs. We applied our method to pangenome graphs built from yeast and human chromosome collections, and demonstrate that our method effectively characterizes discordances between pangenome graph construction methods and scales to real datasets.Availability and implementationpancat compare is published as free Rust software under the AGPL3.0 open source license. Source code and documentation are available at https://github.com/dubssieg/rs-pancat-compare. Snapshot available on Software Heritage at swh:1:dir:61acda8ba3dac1709ed60530147d3871831be629.

To address the issues of long cycles and high costs in the traditional exhibition hall layout design, we propose a method of integrating rule-based and data-driven automatic exhibition space planning and layout generation. Firstly, the specified space is divided according to the exhibition hall design rules, and the space is optimized considering beautification and connection. Then, the visual saliency of the exhibition hall space is calculated, and the position and arrangement of exhibits are selected using a priority queue. Next, using the data generated by rule-based methods as a dataset, and the layout search results from the regional center as input, we achieve the diverse generation by combining semantic prediction network and topological prediction network. Finally, through the prediction model, the optimal placement positions of exhibits are determined in decreasing order of importance. On the exhibition hall dataset, we perform comparative experiments with other methods, including HouseGAN++, Deeplayout, and DCGAN. The results indicate that the proposed method has higher visual similarity, with an average reduction of 22.64% in graph edit distance, and an average increase of 25.33% in connectivity, validating the effectiveness of this method.

Representing molecules as graphs is a natural approach for capturing their structural information, with atoms depicted as nodes and bonds as edges. Although graph-based similarity calculation approaches, such as the graph edit distance, have been proposed for calculating molecular similarity, these approaches are nondeterministic polynomial (NP)-hard and thus computationally infeasible for routine use, unlike fingerprint-based methods. To address this limitation, we developed GESim, an ultrafast graph-based method for calculating molecular similarity on the basis of von Neumann graph entropy. GESim enables molecular similarity calculations by considering entire molecular graphs, and evaluations using two benchmarks for molecular similarity suggest that GESim has the ability to differentiate between highly similar molecules, even in cases where other methods fail to effectively distinguish their similarity. GESim is provided as an open-source package on GitHub at https://github.com/LazyShion/GESim.

Graph similarity computation (GSC) is to calculate the similarity between one pair of graphs, which is a fundamental problem with fruitful applications in the graph community. In GSC, graph edit distance (GED) and maximum common subgraph (MCS) are the two most adopted similarity metrics, both of which are NP-hard to compute. Instead of calculating the exact values, state-of-the-art solutions resort to leveraging graph neural networks (GNNs) to learn data-driven models for the estimation of GED and MCS. Most of them are built on components involving node-level interactions crossing graphs, which engender vast computation overhead but are of little avail in effectiveness. Motivated by this, in the paper, we present GraSP, a simple yet effective GSC approach for GED and MCS prediction. More concretely, GraSP achieves high result efficacy through several key instruments: enhanced node features via positional encoding and a GNN model augmented by a gating mechanism, residual connections, as well as multi-scale pooling. Theoretically, GraSP can surpass the 1-WL test, indicating its high expressiveness. Empirically, extensive experiments comparing GraSP against 10 competitors on multiple widely adopted benchmark datasets showcase the superiority of GraSP over prior arts in terms of both effectiveness and efficiency.

Given a graph pair (G 1 , G 2 ), graph edit distance (GED) is defined as the minimum number of edit operations converting G 1 to G 2 . GED is a fundamental operation widely used in many applications, but its exact computation is NP-hard, so the approximation of GED has gained a lot of attention. Data-driven learning-based methods have been found to provide superior results compared to classical approximate algorithms, but they directly fit the coupling relationship between a pair of vertices from their vertex features. We argue that while pairwise vertex features can capture the coupling cost (discrepancy) of a pair of vertices, the vertex coupling matrix should be derived from the vertex-pair cost matrix through a more well-established method that is aware of the global context of the graph pair, such as optimal transport. In this paper, we propose an ensemble approach that integrates a supervised learning-based method and an unsupervised method, both based on optimal transport. Our learning method, GEDIOT, is based on inverse optimal transport that leverages a learnable Sinkhorn algorithm to generate the coupling matrix. Our unsupervised method, GEDGW, models GED computation as a linear combination of optimal transport and its variant, Gromov-Wasserstein discrepancy, for node and edge operations, respectively, which can be solved efficiently without needing the ground truth. Our ensemble method, GEDHOT, combines GEDIOT and GEDGW to further boost the performance. Extensive experiments demonstrate that our methods significantly outperform the existing methods in terms of the performance of GED computation, edit path generation, and model generalizability.

With the rapid growth of online educational resources, existing personalized course recommendation systems face challenges in multimodal feature integration and limited recommendation interpretability when dealing with complex and diverse instructional content. This paper proposes a graph-enhanced multimodal recommendation method (GEMRec), which effectively integrates text, video, and audio features through a graph attention network and differentiable pooling. Innovatively, GEMRec introduces graph edit distance into the recommendation system to measure the structural similarity between a learner’s knowledge state and course content at the knowledge graph level. Additionally, it combines SHAP (SHapley Additive exPlanations) value computation with large language models to generate reliable and personalized recommendation explanations. Experiments on the MOOCCubeX dataset demonstrate that the GEMRec model exhibits strong convergence and generalization during training. Compared with existing methods, GEMRec achieves 0.267, 0.265, and 0.297 on the Precision@10, Recall@10, and NDCG@10 metrics, respectively, significantly outperforming traditional collaborative filtering and other deep learning models. These results validate the effectiveness of multimodal feature integration and knowledge graph enhancement in improving recommendation performance.

We solve, in the typical-case sense, the multi-sender linearly-decomposable distributed computing problem introduced by tessellated distributed computing. We model real-valued encoders/decoders and demand matrices, and assess structural fidelity via a thresholded graph edit distance between the demand support and the two-hop support of the computed product. Our analysis yields: a closed-form second-moment (Frobenius) risk under spike-and-slab ensembles; deterministic links between thresholded GED and norm error; a Gaussian surrogate with sub-exponential tails that exposes explicit recall lines; concentration of GED and operator-norm control; and a compute-capped design with a visible knee. We map the rules to aeronautical and satellite networks.