Graph-based Deep Learning Research Articles

Spatially resolved subcellular protein-protein interactomics in drug-perturbed lung-cancer cultures and tissues.

Protein-protein interactions (PPIs) regulate signalling pathways and cell phenotypes, and the visualization of spatially resolved dynamics of PPIs would thus shed light on the activation and crosstalk of signalling networks. Here we report a method that leverages a sequential proximity ligation assay for the multiplexed profiling of PPIs with up to 47 proteins involved in multisignalling crosstalk pathways. We applied the method, followed by conventional immunofluorescence, to cell cultures and tissues of non-small-cell lung cancers with a mutated epidermal growth-factor receptor to determine the co-localization of PPIs in subcellular volumes and to reconstruct changes in the subcellular distributions of PPIs in response to perturbations by the tyrosine kinase inhibitor osimertinib. We also show that a graph convolutional network encoding spatially resolved PPIs can accurately predict the cell-treatment status of single cells. Multiplexed proximity ligation assays aided by graph-based deep learning can provide insights into the subcellular organization of PPIs towards the design of drugs for targeting the protein interactome.

CloneRipples: predicting change propagation between code clone instances by graph-based deep learning

CloneRipples: predicting change propagation between code clone instances by graph-based deep learning

PASCL: supervised contrastive learning with perturbative augmentation for particle decay reconstruction

Abstract In high-energy physics, particles produced in collision events decay in a format of a hierarchical tree structure, where only the final decay products can be observed using detectors. However, the large combinatorial space of possible tree structures makes it challenging to recover the actual decay process given a set of final particles. To better analyse the hierarchical tree structure, we propose a graph-based deep learning model to infer the tree structure to reconstruct collision events. In particular, we use a compact matrix representation termed as lowest common ancestor generations matrix, to encode the particle decay tree structure. Then, we introduce a perturbative augmentation technique applied to node features, aiming to mimic experimental uncertainties and increase data diversity. We further propose a supervised graph contrastive learning algorithm to utilize the information of inter-particle relations from multiple decay processes. Extensive experiments show that our proposed supervised graph contrastive learning with perturbative augmentation method outperforms state-of-the-art baseline models on an existing physics-based dataset, significantly improving the reconstruction accuracy. This method provides a more effective training strategy for models with the same parameters and makes way for more accurate and efficient high-energy particle physics data analysis.

Graph-Based Spatial Proximity of Super-Resolved Protein-Protein Interactions Predicts Cancer Drug Responses in Single Cells.

Current bulk molecular assays fail to capture spatial signaling activities in cancers, limiting our understanding of drug resistance mechanisms. We developed a graph-based super-resolution protein-protein interaction (GSR-PPI) technique to spatially resolve single-cell signaling networks and evaluate whether higher resolution microscopy enhances the biological study of PPIs using deep learning classification models. Single-cell spatial proximity ligation assays (PLA, ≤ 9 PPI pairs) were conducted on EGFR mutant (EGFRm) PC9 and HCC827 cells (>10,000 cells) treated with 100nM Osimertinib. Multiplexed PPI images were obtained using wide-field and super-resolution microscopy (Zeiss Airyscan, SRRF). Graph-based deep learning models analyzed subcellular protein interactions to classify drug treatment states and test GSR-PPI on clinical tissue samples. GSR-PPI triangulated PPI nodes into 3D relationships, predicting drug treatment labels. Biological discriminative ability (BDA) was evaluated using accuracy, AUC, and F1 scores. The method was also applied to 3D spatial proteomic molecular pixelation (PixelGen) data from T cells. GSR-PPI outperformed baseline models in predicting drug responses from multiplexed PPI imaging in EGFRm cells. Super-resolution data significantly improved accuracy over localized wide-field imaging. GSR-PPI classified drug treatment states in cancer cells and human lung tissues, with performance improving as imaging resolution increased. It differentiated single and combination drug therapies in HCC827 cells and human tissues. Additionally, GSR-PPI accurately distinguished T-cell stimulation states, identifying key nodes such as CD44, CD45, and CD54. The GSR-PPI framework provides valuable insights into spatial protein interactions and drug responses, enhancing the study of signaling biology and drug resistance. The online version contains supplementary material available at 10.1007/s12195-024-00822-1.

Spatio-temporal Dual Graph Neural Networks for Travel Time Estimation

Travel time estimation is one of the core tasks for the development of intelligent transportation systems. Most previous works model the road segments or intersections separately by learning their spatio-temporal characteristics to estimate travel time. However, due to the continuous alternations of the road segments and intersections in a path, the dynamic features are supposed to be coupled and interactive. Therefore, modeling one of them limits further improvement in accuracy of estimating travel time. To address the above problems, a novel graph-based deep learning framework for travel time estimation is proposed in this article, namely, Spatio-temporal Dual Graph Neural Networks (STDGNN). Specifically, we first establish the node-wise and edge-wise graphs to, respectively, characterize the adjacency relations of intersections and that of road segments. To extract the joint spatio-temporal correlations of the intersections and road segments, we adopt the spatio-temporal dual graph learning approach that incorporates multiple spatial-temporal dual graph learning modules with multi-scale network architectures for capturing multi-level spatial-temporal information from the dual graph. Finally, we employ the multi-task learning approach to estimate the travel time of a given whole route, each road segment and intersection simultaneously. We conduct extensive experiments to evaluate our proposed model on three real-world trajectory datasets, and the experimental results show that STDGNN significantly outperforms several state-of-art baselines.

SpaGIC: graph-informed clustering in spatial transcriptomics via self-supervised contrastive learning.

Spatial transcriptomics technologies enable the generation of gene expression profiles while preserving spatial context, providing the potential for in-depth understanding of spatial-specific tissue heterogeneity. Leveraging gene and spatial data effectively is fundamental to accurately identifying spatial domains in spatial transcriptomics analysis. However, many existing methods have not yet fully exploited the local neighborhood details within spatial information. To address this issue, we introduce SpaGIC, a novel graph-based deep learning framework integrating graph convolutional networks and self-supervised contrastive learning techniques. SpaGIC learns meaningful latent embeddings of spots by maximizing both edge-wise and local neighborhood-wise mutual information of graph structures, as well as minimizing the embedding distance between spatially adjacent spots. We evaluated SpaGIC on seven spatial transcriptomics datasets across various technology platforms. The experimental results demonstrated that SpaGIC consistently outperformed existing state-of-the-art methods in several tasks, such as spatial domain identification, data denoising, visualization, and trajectory inference. Additionally, SpaGIC is capable of performing joint analyses of multiple slices, further underscoring its versatility and effectiveness in spatial transcriptomics research.

CGDGMDA-Net: discovering microbe-disease and drug associations through CTGAN and graph-based deep learning

CGDGMDA-Net: discovering microbe-disease and drug associations through CTGAN and graph-based deep learning

From Static to Dynamic Structures: Improving Binding Affinity Prediction with Graph-Based Deep Learning.

Accurate prediction of protein-ligand binding affinities is an essential challenge in structure-based drug design. Despite recent advances in data-driven methods for affinity prediction, their accuracy is still limited, partially because they only take advantage of static crystal structures while the actual binding affinities are generally determined by the thermodynamic ensembles between proteins and ligands. One effective way to approximate such a thermodynamic ensemble is to use molecular dynamics (MD) simulation. Here, an MD dataset containing 3,218 different protein-ligand complexes is curated, and Dynaformer, a graph-based deep learning model is further developed to predict the binding affinities by learning the geometric characteristics of the protein-ligand interactions from the MD trajectories. In silico experiments demonstrated that the model exhibits state-of-the-art scoring and ranking power on the CASF-2016 benchmark dataset, outperforming the methods hitherto reported. Moreover, in a virtual screening on heat shock protein 90 (HSP90) using Dynaformer, 20 candidates are identified and their binding affinities are further experimentally validated. Dynaformer displayed promising results in virtual drug screening, revealing 12 hit compounds (two are in the submicromolar range), including several novel scaffolds. Overall, these results demonstrated that the approach offer a promising avenue for accelerating the early drug discoveryprocess.

A graph neural network-based teammate recommendation model for knowledge-intensive crowdsourcing

Current research on teammate recommendation decision in various fields is difficult to apply to knowledge-intensive crowdsourcing, which needs to make a trade-off between solver willingness and team performance. To weigh the bilateral benefits between the solver and the seeker, this paper proposes a teammate recommendation model based on multi-criteria decision-making of solver willingness and team performance. The prediction of willingness to team up is driven by explicit competitiveness, teammate familiarity, and social capital. By adapting the graph-based deep learning model, we transform the solver performance prediction into a bipartite network link prediction problem. We jointly compute the solver willingness and team performance to make the final recommendation decisions. The experimental results based on Kaggle data prove the effectiveness of our method. The findings provide managerial implications and a practical reference for the design and operation of teammate recommender systems.

Harmonic Wavelet Neural Network for Discovering Neuropathological Propagation Patterns in Alzheimer's Disease.

Emerging research indicates that the degenerative biomarkers associated with Alzheimer's disease (AD) exhibit a non-random distribution within the cerebral cortex, instead following the structural brain network. The alterations in brain networks occur much earlier than the onset of clinical symptoms, thereby affecting the progression of brain disease. In this context, the utilization of computational methods to ascertain the propagation patterns of neuropathological events would contribute to the comprehension of the pathophysiological mechanism involved in the evolution of AD. Despite the encouraging findings achieved by existing graph-based deep learning approaches in analyzing irregular graph data, their applications in identifying the spreading pathway of neuropathology are limited due to two disadvantages. They include (1) lack of a common brain network as an unbiased reference basis for group comparison, and (2) lack of an appropriate mechanism for the identification of propagation patterns. To this end, we propose a proof-of-concept harmonic wavelet neural network (HWNN) to predict the early stage of AD and localize disease-related significant wavelets, which can be used to characterize the spreading pathways of neuropathological events across the brain network. The extensive experiments constructed on both synthetic and real datasets demonstrate that our proposed method achieves superior performance in classification accuracy and statistical power of identifying propagation patterns, compared with other representative approaches.

DÉDUCTION EFFICACE DE LANGUE OFFENSIVE UTILISANT L'APPRENTISSAGE PROFONDE DANS LES MÉDIAS SOCIAUX

Offensive language detection is the technique of identifying and detecting user-generated offensive comments such as insults, pain, profanity, and racism that are targeted at a specific individual or group on social media. As social media platforms become more prominent, offensive language is used more frequently, becoming a major challenge in modern society. A novel effective offensive language classification (EOLC) technique has been proposed to overcome these challenges. English language tweets from YouTube and X (Twitter) with offensive, mild, swear, and non-offensive tweets are used in this paper. Initially, the tweets and comments are pre-processed, and the features are extracted using different techniques, namely term frequency-inverse document frequency (TF-IDF), WordVec, and lexicon-based features. The extracted features are classified using the graph-based deep learning (GDL) method for numerical representation and decision-making. GDL network is optimized with red fox optimization (RFO) to normalize the weight and biases of the network and achieve better accuracy. The proposed GDL model achieves the highest levels of classification accuracy on the X (Twitter) and YouTube datasets, with 95.5 % and 96.8 %, respectively. The results obtained from GDL are more accurate and of higher quality than those obtained from traditional classifiers. The proposed EOLC method improves the overall accuracy by 5.56 %, 7.4 %, 7.7 %, and 10.2 % better than Text CNN, CNN-LSTM, DRNN, and LogitBoost, respectively.

ToxMPNN: A deep learning model for small molecule toxicity prediction.

Machine learning (ML) has shown a great promise in predicting toxicity of small molecules. However, the availability of data for such predictions is often limited. Because of the unsatisfactory performance of models trained on a single toxicity endpoint, we collected toxic small molecules with multiple toxicity endpoints from previous study. The dataset comprises 27 toxic endpoints categorized into seven toxicity classes, namely, carcinogenicity and mutagenicity, acute oral toxicity, respiratory toxicity, irritation and corrosion, cardiotoxicity, CYP450, and endocrine disruption. In addition, a binary classification Common-Toxicity task was added based on the aforementioned dataset. To improve the performance of the models, we added marketed drugs as negative samples. This study presents a toxicity predictive model, ToxMPNN, based on the message passing neural network (MPNN) architecture, aiming to predict the toxicity of small molecules. The results demonstrate that ToxMPNN outperforms other models in capturing toxic features within the molecular structure, resulting in more precise predictions with the ROC_AUC testing score of 0.886 for the Toxicity_drug dataset. Furthermore, it was observed that adding marketed drugs as negative samples not only improves the predictive performance of the binary classification Common-Toxicity task but also enhances the stability of the model prediction. It shows that the graph-based deep learning (DL) algorithms in this study can be used as a trustworthy and effective tool to assess small molecule toxicity in the development of new drugs.

Toxicity prediction of chemicals using OECD test guideline data with graph-based deep learning models

Toxicity prediction of chemicals using OECD test guideline data with graph-based deep learning models

Towards high-accuracy axial springback: Mesh-based simulation of metal tube bending via geometry/process-integrated graph neural networks

Springback has always been a stubborn defect that affects the axial accuracy of metal bending. The finite element simulation of springback enables effective control and precise compensation to improve the forming quality. Affected by the material, size, and forming process, the generation pattern of the springback defect remains further exploration. Graph-based deep learning techniques support differentiable high-dimensional physics simulations with significantly lower computational resources and comparable accuracy. In this paper, a framework based on geometry/process-integrated Graph Neural Networks (GNN) is presented for simulating axial springback of mesh-based metal bending, with a specific focus on the tube bending. The framework adopts an encode-process-decode structure, equipped with parallel graph network blocks and data augmentation strategy, to capture the axial springback information with high fidelity in a form of mesh. Comparative experiments reveal that our framework achieves an accuracy comparable to that of finite elements method, yielding an average nodal error below 1 mm and outperforming three representative GNN baselines. The framework remarkably enhances the simulation efficiency, exhibiting a four-order-of-magnitude improvement on CPU and a six-order-of-magnitude improvement on GPU compared to finite elements method. Furthermore, we successfully apply the proposed framework to simulate the springback of plate V-bending, showcasing its robust generalization ability. These results illustrate the capability of our GNN framework to achieve the accurate and real-time springback simulation, with significant implications for digital twin development and bent tube quality optimization.

Mitochondrial nanoprobe for precise cellular and drug analysis via graph Neural network

Mitochondrial morphology is crucial for cell identification and drug toxicity evaluation, yet it is challenged by mitochondrial complexities and limitations in imaging technologies. We develop an ultrasmall (∼1.5 nm) fluorescent carbon dot (CD), specifically designed to target mitochondria in live cells. These CDs exhibit near-infrared fluorescence at 685 nm, enhancing fluorescence imaging’s signal-to-noise ratio by 2–3 times. Their dense-blinking behavior enables rapid fluctuation-based super-resolution imaging with as few as 10 frames, thereby allowing for detailed visualization of mitochondrial morphology and dynamics. We further propose a graph-based deep learning framework that integrates multidimensional mitochondrial features, which are updated using a Graph Neural Network (GNN), to achieve precise mitochondria-based cellular typing and drug toxicity analysis with up to 98% accuracy. Additionally, we pioneer the use of interpretability algorithms to elucidate the GNN model, revealing how the depicted mitochondrial features by the CDs drive these predictions. This approach has significant implications in cellular and toxicological research, offering a unique tool for deciphering cellular behaviors and drug interactions.

DDMut-PPI: predicting effects of mutations on protein-protein interactions using graph-based deep learning.

Protein-protein interactions (PPIs) play a vital role in cellular functions and are essential for therapeutic development and understanding diseases. However, current predictive tools often struggle to balance efficiency and precision in predicting the effects of mutations on these complex interactions. To address this, we present DDMut-PPI, a deep learning model that efficiently and accurately predicts changes in PPI binding free energy upon single and multiple point mutations. Building on the robust Siamese network architecture with graph-based signatures from our prior work, DDMut, the DDMut-PPI model was enhanced with a graph convolutional network operated on the protein interaction interface. We used residue-specific embeddings from ProtT5 protein language model as node features, and a variety of molecular interactions as edge features. By integrating evolutionary context with spatial information, this framework enables DDMut-PPI to achieve a robust Pearson correlation of up to 0.75 (root mean squared error: 1.33 kcal/mol) in our evaluations, outperforming most existing methods. Importantly, the model demonstrated consistent performance across mutations that increase or decrease binding affinity. DDMut-PPI offers a significant advancement in the field and will serve as a valuable tool for researchers probing the complexities of protein interactions. DDMut-PPI is freely available as a web server and an application programming interface at https://biosig.lab.uq.edu.au/ddmut_ppi.

Factor graph-based deep reinforcement learning for path selection scheme in full-duplex wireless multihop networks

A wireless multihop network (WMN) is set of wirelessly connected nodes without an aid of centralized infrastructure that can forward any packets via intermediate nodes by a multihop fashion. In the WMN, there are still some issues that need to be resolved, like due to any source node may choose an uncertainty path to send their packets through the multihop fashion and this leads to the performance of network capacity can degrade drastically. To solve this problem, in this research, we propose two novel path selection algorithms called SNR-based learning path selection (NLPS) algorithm and SINR-based learning path selection (INLPS) algorithm, which are incorporated with the deep reinforcement learning (DRL) to select the best multihop path from any source node to a destination node with highest end-to-end (E2E) throughput. Besides that, a factor graph (FG) approach and a nested lattice code (NLC) representation are used to reduce the computation time. According to the numerical studies with the NLC is applied, our simulation results reveal that the proposed NLPS and INLPS algorithms can improve the overall average network capacity up to 3.1 times and 10.5 times compared to FG, respectively. However, the overall average computation time are highly increased for NLPS and INLPS, i.e., about 0.627 s and 1.221 s, respectively compared to FG, which is about 0.006 s. In other words, both NLPS and INLPS algorithms can achieve high network capacity and moderate computation time.

Intelligent Financial Risk Warning for Enterprises Through Knowledge Graph-Based Deep Learning

The financial risk warning methods for enterprises have always been a practical concern. In digital society, the computational intelligence has brought more spirit to this demand. This paper first introduces the current situation of the development of Knowledge graph technology, describes the deep learning fusion method based on Knowledge graph, and expresses the feasibility of this study. Then, according to the requirements of Knowledge graph, it completes the method fusion of core data training and extraction, and completes the adaptive deep learning design for the Beautiful SCOP database, and establishes a STDE-FG financial risk early warning model. Through empirical analysis, the shortcomings of this model were identified, and a comparison of optimized and optimized results was completed. Two aspects of phenomenon can be found from experimental results. For one thing, the accuracy of the unoptimized STDE-FG early warning model has been improved by 37.5–55.3% compared to traditional prediction models, but the prediction value during legal person changes has a greater error than traditional prediction values. For another, the optimized STDE-FG early warning model has also improved its accuracy in predicting new investments and equity changes, with improvements of 17–32% and 16–28%, respectively, with significant changes. This model will have a positive impact on improving enterprise risk management capabilities and reducing financial risk costs.

GTADC: A Graph-Based Method for Inferring Cell Spatial Distribution in Cancer Tissues.

The heterogeneity of tumors poses a challenge for understanding cell interactions and constructing complex ecosystems within cancer tissues. Current research strategies integrate spatial transcriptomics (ST) and single-cell sequencing (scRNA-seq) data to thoroughly analyze this intricate system. However, traditional deep learning methods using scRNA-seq data tend to filter differentially expressed genes through statistical methods. In the context of cancer tissues, where cancer cells exhibit significant differences in gene expression compared to normal cells, this heterogeneity renders traditional analysis methods incapable of accurately capturing differences between cell types. Therefore, we propose a graph-based deep learning method, GTADC, which utilizes Silhouette scores to precisely capture genes with significant expression differences within each cell type, enhancing the accuracy of gene selection. Compared to traditional methods, GTADC not only considers the expression similarity of genes within their respective clusters but also comprehensively leverages information from the overall clustering structure. The introduction of graph structure effectively captures spatial relationships and topological structures between the two types of data, enabling GTADC to more accurately and comprehensively resolve the spatial composition of different cell types within tissues. This refinement allows GTADC to intricately reconstruct the cellular spatial composition, offering a precise solution for inferring cell spatial composition. This method allows for early detection of potential cancer cell regions within tissues, assessing their quantity and spatial information in cell populations. We aim to achieve a preliminary estimation of cancer occurrence and development, contributing to a deeper understanding of early-stage cancer and providing potential support for early cancer diagnosis.

Dual-Graph Attention Convolution Network for 3-D Point Cloud Classification.

Three-dimensional point cloud classification is fundamental but still challenging in 3-D vision. Existing graph-based deep learning methods fail to learn both low-level extrinsic and high-level intrinsic features together. These two levels of features are critical to improving classification accuracy. To this end, we propose a dual-graph attention convolution network (DGACN). The idea of DGACN is to use two types of graph attention convolution operations with a feedback graph feature fusion mechanism. Specifically, we exploit graph geometric attention convolution to capture low-level extrinsic features in 3-D space. Furthermore, we apply graph embedding attention convolution to learn multiscale low-level extrinsic and high-level intrinsic fused graph features together. Moreover, the points belonging to different parts in real-world 3-D point cloud objects are distinguished, which results in more robust performance for 3-D point cloud classification tasks than other competitive methods, in practice. Our extensive experimental results show that the proposed network achieves state-of-the-art performance on both the synthetic ModelNet40 and real-world ScanObjectNN datasets.