Graph Convolutional Neural Networks Research Articles

EHR-HGCN: An Enhanced Hybrid Approach for Text Classification Using Heterogeneous Graph Convolutional Networks in Electronic Health Records.

Text classification is a central part of natural language processing, with important applications in understanding the knowledge behind biomedical texts including electronic health records (EHR). In this article, we propose a novel heterogeneous graph convolutional network method for classifying EHR texts. Our method, called EHR-HGCN, is able to combine context-sensitive word and sentence embeddings with structural sentence-level and word-level relation information to perform text classification. EHR-HGCN reframes EHR text classification as a graph classification task to better capture structural information about the document using a heterogeneous graph. To mine contextual information from a document, EHR-HGCN first applies a bidirectional recurrent neural network (BiRNN) on word embeddings obtained via Global Vectors for word representation (GloVe) to obtain context-sensitive word-level and sentence-level embeddings. To mine structural relationships from the document, EHR-HGCN then constructs a heterogeneous graph over the word and sentence embeddings, where sentence-word and word-word relationships are represented by graph edges. Finally, a heterogeneous graph convolutional neural network is used to classify documents by their graph representation. We evaluate EHR-HGCN on a variety of standard text classification benchmarks and find that EHR-HGCN has higher accuracy and F1-score than other representative machine learning and deep learning methods. We also apply EHR-HGCN to the MedLit benchmark and find it performs with high accuracy and F1-score on the task of section classification in EHR texts. Our ablation experiments show that the heterogeneous graph construction and heterogeneous graph convolutional network are critical to the performance of EHR-HGCN.

Human-Robot Collaboration Through a Multi-Scale Graph Convolution Neural Network With Temporal Attention

Human-Robot Collaboration Through a Multi-Scale Graph Convolution Neural Network With Temporal Attention

GCN-Based Risk Prediction for Necrosis Slide of Hepatocellular Carcinoma.

Hepatocellular carcinoma (HCC) is one of the most common cancers in the world which ranks fourth in cancer deaths. Primary pathological necrosis is an effective prognostic indicator for hepatocellular carcinoma. We propose a GCN-based approach that mimics the pathologist's perspective for global assessment of necrosis tissue distribution to analyze patient survival. Specifically, we introduced a graph convolutional neural network to construct a spatial map with necrotic tissue and tumor tissue as graph nodes, aiming to mine the contextual information between necrotic tissue in pathological sections. We used 1381 slides from 303 patients from the First Affiliated Hospital of Zhejiang University School to train the model and used TCGA-LIHC for external validation. The C-index of our method outperforms the baseline by about 4.45%, which proves that the information about the spatial distribution of necrosis learned by GCN is meaningful for guiding patient prognosis.

Accelerating defect predictions in semiconductors using graph neural networks

First-principles computations reliably predict the energetics of point defects in semiconductors but are constrained by the expense of using large supercells and advanced levels of theory. Machine learning models trained on computational data, especially ones that sufficiently encode defect coordination environments, can be used to accelerate defect predictions. Here, we develop a framework for the prediction and screening of native defects and functional impurities in a chemical space of group IV, III–V, and II–VI zinc blende semiconductors, powered by crystal Graph-based Neural Networks (GNNs) trained on high-throughput density functional theory (DFT) data. Using an innovative approach of sampling partially optimized defect configurations from DFT calculations, we generate one of the largest computational defect datasets to date, containing many types of vacancies, self-interstitials, anti-site substitutions, impurity interstitials and substitutions, as well as some defect complexes. We applied three types of established GNN techniques, namely crystal graph convolutional neural network, materials graph network, and Atomistic Line Graph Neural Network (ALIGNN), to rigorously train models for predicting defect formation energy (DFE) in multiple charge states and chemical potential conditions. We find that ALIGNN yields the best DFE predictions with root mean square errors around 0.3 eV, which represents a prediction accuracy of 98% given the range of values within the dataset, improving significantly on the state-of-the-art. We further show that GNN-based defective structure optimization can take us close to DFT-optimized geometries at a fraction of the cost of full DFT. The current models are based on the semi-local generalized gradient approximation-Perdew–Burke–Ernzerhof (PBE) functional but are highly promising because of the correlation of computed energetics and defect levels with higher levels of theory and experimental data, the accuracy and necessity of discovering novel metastable and low energy defect structures at the PBE level of theory before advanced methods could be applied, and the ability to train multi-fidelity models in the future with new data from non-local functionals. The DFT-GNN models enable prediction and screening across thousands of hypothetical defects based on both unoptimized and partially optimized defective structures, helping identify electronically active defects in technologically important semiconductors.

Semisupervised Vector Quantization in Visual SLAM Using HGCN

We present a novel vector quantization (VQ) module for the two state-of-the-art long-range simultaneous localization and mapping (SLAM) algorithms. The VQ task in SLAM is generally performed using unsupervised methods. We provide an alternative approach trough embedding a semisupervised hyperbolic graph convolutional neural network (HGCN) in the VQ step of the SLAM processes. The SLAM platforms we have utilized for this purpose are fast appearance-based mapping (FABMAP) and oriented fast and rotated short (ORB), both of which rely on extracting the features of the captured images in their loop closure detection (LCD) module. For the first time, we have considered the space formed by these SURF features, robust image descriptors, as a graph, enabling us to apply an HGCN in the VQ section which results in an improved LCD performance. The HGCN vector quantizes the SURF feature space, leading to a bag-of-word (BoW) representation construction of the images. This representation is subsequently used to determine LCD accuracy and recall. Our approaches in this study are referred to as HGCN-FABMAP and HGCN-ORB. The main advantage of using HGCN in the LCD section is that it scales linearly when the features are accumulated. The benchmarking experiments show the superiority of our methods in terms of both trajectory generation accuracy in small-scale paths and LCD accuracy and recall for large-scale problems.

A heterogeneous 3-D stacked PIM accelerator for GCN-based recommender systems

AbstractModern recommendation systems integrate graph convolution neural networks (GCN) for enhancing embedding representation. Compared with widely deployed neural network-based models, the extra message propagation layer of GCN-based recommendation is featured with extensive computations and irregular memory access. However, architecture designs for prevailing deep neural network recommendation models assume simple pooling in the embedding layer. ReRAM-based GCN accelerators are specialized for graph-related operations. However, they are designed for general graphs, while GCN-based recommendation models mainly operate on the user-item graph. In this paper, we proposed a resistive random accessed memory (ReRAM) based processing-in-memory (PIM) accelerator, ReGCNR, for GCN-based recommendation. ReGCNR is featured with three key innovations. First, we exploit the 3-dimensional (3-D) stacked heterogeneous ReRAM to fit with the large-size embedding table and user-item graph. Then, we propose a joint degree mapping schema that maximizes the efficiency of the execution pipeline. After that, ReGCNR assembles a well-coordinated pipeline and hardware scheduling design to boost overall system performance. Results show that ReGCNR outperforms GPU by 69.83$$\times$$ × and 56.67$$\times$$ × in terms of average speedup and energy saving, respectively. In addition, ReGCNR outperforms state-of-the-art ReRAM-based solutions by 11.13$$\times$$ × speedups and 7.22$$\times$$ × energy savings on average.

A multi-type vulnerability detection framework with parallel perspective fusion and hierarchical feature enhancement

A core problem of vulnerability detection is to detect multi-type vulnerabilities simultaneously by characterizing vulnerabilities of high diversity and complexity in real program source code. Current methods mainly adjust and compromise multiple code representations such as code sequence and code graph based on composite graph. However, sequential features extracted by graph are hardly sufficient to model the contextual semantic associations of the token sequence. Meanwhile, structural features of the code graph extracted by models based on Euclidean Graph Neural Network are difficult to fit the tree-like calling relationships between code lines. These limitations make it difficult to detect diverse vulnerabilities. In addition, most of the existing models ignore the type of code statement, which is closely associated with some specific vulnerability types. In this paper, we propose a Parallelism Framework with Hierarchical feature Enhancement for Multi-type Vulnerability Detection (PFHE-MVD). PFHE-MVD models program code from three parallel perspectives, containing sequence, code graph, and Abstract Syntax Tree statistic. Hyperbolic Graph Convolutional Neural Network is integrated to model the top-down hierarchical calling structure in program code graph through hyperbolic space mapping. Besides, the statement type of code is embedded along with the code text to strengthen the identification ability for different types of vulnerabilities. Experimental results show that PFHE-MVD achieves new state-of-the-art results in multi-type vulnerability detection. PFHE-MVD captures tree-like hierarchical code structure feature and enhances the distinguishing ability for vulnerabilities by code statement type embedding.

Graph Convolutional Neural Network for Pressure Prediction in Water Distribution Network Sites

Graph Convolutional Neural Network for Pressure Prediction in Water Distribution Network Sites

DeepMCGCN: Multi-channel Deep Graph Neural Networks

Graph neural networks (GNNs) have shown powerful capabilities in modeling and representing graph structural data across various graph learning tasks as an emerging deep learning approach. However, most existing GNNs focus on single-relational graphs and fail to fully utilize the rich and diverse relational information present in real-world graph data. In addition, deeper GNNs tend to suffer from overfitting and oversmoothing issues, leading to degraded model performance. To deeply excavate the multi-relational features in graph data and strengthen the modeling and representation abilities of GNNs, this paper proposes a multi-channel deep graph convolutional neural network method called DeepMCGCN. It constructs multiple relational subgraphs and adopts multiple GCN channels to learn the characteristics of different relational subgraphs separately. Cross-channel connections are utilized to obtain interactions between different relational subgraphs, which can learn node embeddings richer and more discriminative than single-channel GNNs. Meanwhile, it alleviates overfitting issues of deep models by optimizing convolution functions and adding residual connections between and within channels. The DeepMCGCN method is evaluated on three real-world datasets, and the experimental results show that its node classification performance outperforms that of single-channel GCN and other benchmark models, which improves the modeling and representation capabilities of the model.

Enhanced model iteration algorithm with graph neural network for diffuse optical tomography.

Diffuse optical tomography (DOT) employs near-infrared light to reveal the optical parameters of biological tissues. Due to the strong scattering of photons in tissues and the limited surface measurements, DOT reconstruction is severely ill-posed. The Levenberg-Marquardt (LM) is a popular iteration method for DOT, however, it is computationally expensive and its reconstruction accuracy needs improvement. In this study, we propose a neural model based iteration algorithm which combines the graph neural network with Levenberg-Marquardt (GNNLM), which utilizes a graph data structure to represent the finite element mesh. In order to verify the performance of the graph neural network, two GNN variants, namely graph convolutional neural network (GCN) and graph attention neural network (GAT) were employed in the experiments. The results showed that GCNLM performs best in the simulation experiments within the training data distribution. However, GATLM exhibits superior performance in the simulation experiments outside the training data distribution and real experiments with breast-like phantoms. It demonstrated that the GATLM trained with simulation data can generalize well to situations outside the training data distribution without transfer training. This offers the possibility to provide more accurate absorption coefficient distributions in clinical practice.

Spatial-temporal deep learning model based on Similarity Principle for dock shared bicycles ridership prediction

Demand prediction plays a critical role in traffic research. The key challenge of traffic demand prediction lies in modeling the complex spatial dependencies and temporal dynamics. However, there is no mature and widely accepted concept to support the solution of the above challenge. Essentially, a prediction model combined with similar objects in temporal and spatial dimensions could obtain better performance. This paper proposes a concept called the Similarity-based Principle (SP), which is applied to improve the prediction performance of deep learning models in complex traffic scenarios. For the temporal components, the long-term temporal dynamics in contemporaneous historical data for ridership are extracted by the Stacked Autoencoder (SAE) method. For the spatial components, the activity-based spatial geographic information (ABG-information) is used to capture the spatial correlation of the traffic network, which is reflected in the daily activities of humans. Specifically, the SP is applied to a Spatio-temporal Graph Convolutional Neural Network (STGCNN) model. In the case study, the Similarity-based Principle Spatio-temporal Graph Convolutional Neural Network (SP-STGCNN) model predicts demand for bicycle sharing in San Francisco. The results show that the SP effectively improves the model's performance. The prediction accuracy is enhanced by up to 10.34% compared with STGCNN. For spatial relationships, the model using the geographic information attribute performs better than that using the road information attribute and the distance attribute. It is proved that the construction of the Spatio-temporal model-based similarity principle can improve the performance.

AMYGNN: A Graph Convolutional Neural Network-Based Approach for Predicting Amyloid Formation from Polypeptides.

There has been an increasing interest in the use of amyloids for constructing various functional materials. The design of amyloid-associated functional materials requires the identification of the core peptide sequences as the fundamental building block. The existing computational methods are limited in terms of delineating polypeptides, the typical non-Euclidean structural data, and they fail to capture the dynamic interactions between amino acids due to ignoring the contextual information from surrounding amino acids. Here, we first propose the use of a state-of-the-art graph convolutional neural network for predicting the trends of amyloid formation from specific peptide sequences (AMYGNN) by abstracting each polypeptide as a graph, in which the constituting amino acids are viewed as nodes and edges characterizing the connections between pairs of amino acids are established when they meet a given distance threshold (Cα-Cα ≤ 5 Å). Our model achieves high performance with accuracy (0.9208), G-mean (0.9203), MCC (0.8417), and F1 (0.9235) in determining the characteristic peptide sequences to form amyloid. 32 of 534 crucial amino acid properties that greatly contribute to the formation of amyloids are ascertained, and the β-folding-like graph structure of a polypeptide is believed to be essential for the formation of amyloid. Our model enables the mapping of polypeptides with underlying interactions between amino acids and provides a quick and precise predictive framework for directing the construction of amyloid-associated functional materials.

Identification of breast cancer subtypes based on graph convolutional network

Identification of molecular subtypes of malignant tumors plays a vital role in individualized diagnosis, personalized treatment, and prognosis prediction of cancer patients. The continuous improvement of comprehensive tumor genomics database and the ongoing breakthroughs in deep learning technology have driven further advancements in computer-aided tumor classification. Although the existing classification methods based on gene expression omnibus database take the complexity of cancer molecular classification into account, they ignore the internal correlation and synergism of genes. To solve this problem, we propose a multi-layer graph convolutional network model for breast cancer subtype classification combined with hierarchical attention network. This model constructs the graph embedding datasets of patients' genes, and develops a new end-to-end multi-classification model, which can effectively recognize molecular subtypes of breast cancer. A large number of test data prove the good performance of this new model in the classification of breast cancer subtypes. Compared to the original graph convolutional neural networks and two mainstream graph neural network classification algorithms, the new model has remarkable advantages. The accuracy, weight-F1-score, weight-recall, and weight-precision of our model in seven-category classification has reached 0.851 7, 0.823 5, 0.851 7 and 0.793 6 respectively. In the four-category classification, the results are 0.928 5, 0.894 9, 0.928 5 and 0.865 0 respectively. In addition, compared with the latest breast cancer subtype classification algorithms, the method proposed in this paper also achieved the highest classification accuracy. In summary, the model proposed in this paper may serve as an auxiliary diagnostic technology, providing a reliable option for precise classification of breast cancer subtypes in the future and laying the theoretical foundation for computer-aided tumor classification.

Research on Lane-Changing Decision Making and Planning of Autonomous Vehicles Based on GCN and Multi-Segment Polynomial Curve Optimization.

This paper considers the interactive effects between the ego vehicle and other vehicles in a dynamic driving environment and proposes an autonomous vehicle lane-changing behavior decision-making and trajectory planning method based on graph convolutional networks (GCNs) and multi-segment polynomial curve optimization. Firstly, hierarchical modeling is applied to the dynamic driving environment, aggregating the dynamic interaction information of driving scenes in the form of graph-structured data. Graph convolutional neural networks are employed to process interaction information and generate ego vehicle's driving behavior decision commands. Subsequently, collision-free drivable areas are constructed based on the dynamic driving scene information. An optimization-based multi-segment polynomial curve trajectory planning method is employed to solve the optimization model, obtaining collision-free motion trajectories satisfying dynamic constraints and efficiently completing the lane-changing behavior of the vehicle. Finally, simulation and on-road vehicle experiments are conducted for the proposed method. The experimental results demonstrate that the proposed method outperforms traditional decision-making and planning methods, exhibiting good robustness, real-time performance, and strong scenario generalization capabilities.

Reagent dosage inference based on graph convolutional memory perception network for zinc roughing flotation

In froth flotation, the precise presetting of reagent dosage is crucial for enhancing mineral recovery and flotation efficiency. Considering that the manual operation method is susceptible to some subjective factors, many machine vision-based intelligent reagent dosage presetting approaches are developed. However, prevailing presetting methods primarily rely on empirical knowledge, and ignore the potential intelligent reasoning processes, resulting in a suboptimal flotation performance. In this paper, a reagent dosage inference method based on a graph convolutional memory perception network is proposed, which can leverage the power of both empirical knowledge and neural network to deduce the optimal reagent dosages. This method constructs a memory unit stored with representative operation data, which can be served as a reservoir of empirical knowledge, facilitating dosage inference by referencing relevant data according to the current flotation condition. Subsequently, a graph convolutional neural network is developed to acquire the logical connections between reagent dosage and key parameters of the flotation process. Finally, with the guidance of empirical knowledge, a multi-head attention perception module is employed to deduce the optimal reagent dosage according to the current flotation condition and the acquired logical connection. The effectiveness of the proposed method has been validated in the experiment. Compared to other methods, the proposed method obtains more accurate inference results, and its inferred reagent dosage achieves more satisfactory flotation effect.

GCNGAT: Drug–disease association prediction based on graph convolution neural network and graph attention network

Predicting drug–disease associations can contribute to discovering new therapeutic potentials of drugs, and providing important association information for new drug research and development. Many existing drug–disease association prediction methods have not distinguished relevant background information for the same drug targeted to different diseases. Therefore, this paper proposes a drug–disease association prediction model based on graph convolutional network and graph attention network (GCNGAT) to reposition marketed drugs under the distinguishment of background information. Firstly, in order to obtain initial drug–disease information, a drug–disease heterogeneous graph structure is constructed based on all known drug–disease associations. Secondly, based on the heterogeneous graph structure, the corresponding subgraphs of each group of drug–disease association pairs are extracted to distinguish different background information for the same drug from different diseases. Finally, a model combining Graph neural network with global Average pooling (GnnAp) is designed to predict potential drug–disease associations by learning drug–disease interaction feature representations. The experimental results show that adding subgraph extraction can effectively improve the prediction performance of the model, and the graph representation learning module can fully extract the deep features of drug–disease. Using the 5-fold cross-validation, the proposed model (GCNGAT) achieves AUC (Area Under the receiver operating characteristic Curve) values of 0.9182 and 0.9417 on the PREDICT dataset and CDataset dataset, respectively. Compared with other predictors on the same dataset (PREDICT dataset), GCNGAT outperforms the existing best-performing model (PSGCN), with a 1.58% increase in the AUC value. It is anticipated that this model can provide experimental reference for drug repositioning and further promote the drug research and development process.

Prediction of multiple types of drug interactions based on multi-scale fusion and dual-view fusion.

Potential drug-drug interactions (DDI) can lead to adverse drug reactions (ADR), and DDI prediction can help pharmacy researchers detect harmful DDI early. However, existing DDI prediction methods fall short in fully capturing drug information. They typically employ a single-view input, focusing solely on drug features or drug networks. Moreover, they rely exclusively on the final model layer for predictions, overlooking the nuanced information present across various network layers. To address these limitations, we propose a multi-scale dual-view fusion (MSDF) method for DDI prediction. More specifically, MSDF first constructs two views, topological and feature views of drugs, as model inputs. Then a graph convolutional neural network is used to extract the feature representations from each view. On top of that, a multi-scale fusion module integrates information across different graph convolutional layers to create comprehensive drug embeddings. The embeddings from the two views are summed as the final representation for classification. Experiments on two real-world datasets demonstrate that MSDF achieves higher accuracy than state-of-the-art methods, as the dual-view, multi-scale approach better captures drug characteristics.

Transfer learning graph representations of molecules for pKa, 13C-NMR, and solubility

We explore transfer learning models from a pre-trained graph convolutional neural network representation of molecules, obtained from SchNet, to predict 13C-NMR, pKa, and log S solubility. SchNet learns a graph representation of a molecule by associating each atom with an “embedding vector” and interacts the atom-embeddings with each other by leveraging graph convolutional filters on their interatomic distances. We pre-trained SchNet on molecular energy and demonstrate that the pre-trained atomistic embeddings can then be used as a transferable representation for a wide array of properties. On the one hand, for atomic properties such as micro-pK1 and 13C-NMR, we investigate two models, one linear and one neural net, that input pre-trained atom-embeddings of a particular atom (e.g. carbon) and predict a local property (e.g., 13C-NMR). On the other hand, for molecular properties such as solubility, a size-extensive graph model is built using the embeddings of all atoms in the molecule as input. For all cases, qualitatively correct predictions are made with relatively little training data (<1000 training points), showcasing the ease with which pre-trained embeddings pick up on important chemical patterns. The proposed models successfully capture well-understood trends of pK1 and solubility. This study advances our understanding of current neural net graph representations and their capacity for transfer learning applications in chemistry.

Dedicated Hardware Accelerators for Processing of Sparse Matrices and Vectors: A Survey

Performance in scientific and engineering applications such as computational physics, algebraic graph problems or Convolutional Neural Networks (CNN), is dominated by the manipulation of large sparse matrices—matrices with a large number of zero elements. Specialized software using data formats for sparse matrices has been optimized for the main kernels of interest: SpMV and SpMSpM matrix multiplications, but due to the indirect memory accesses, the performance is still limited by the memory hierarchy of conventional computers. Recent work shows that specific hardware accelerators can reduce memory traffic and improve the execution time of sparse matrix multiplication, compared to the best software implementations. The performance of these sparse hardware accelerators depends on the choice of the sparse format, COO , CSR , etc, the algorithm, inner-product , outer-product , Gustavson , and many hardware design choices. In this article, we propose a systematic survey which identifies the design choices of state-of-the-art accelerators for sparse matrix multiplication kernels. We introduce the necessary concepts and then present, compare, and classify the main sparse accelerators in the literature, using consistent notations. Finally, we propose a taxonomy for these accelerators to help future designers make the best choices depending on their objectives.

Human-airway surface mesh smoothing based on graph convolutional neural networks

Background and ObjectiveA detailed representation of the airway geometry in the respiratory system is critical for predicting precise airflow and pressure behaviors in computed tomography (CT)-image-based computational fluid dynamics (CFD). The CT-image-based geometry often contains artifacts, noise, and discontinuities due to the so-called stair step effect. Hence, an advanced surface smoothing is necessary. The existing smoothing methods based on the Laplacian operator drastically shrink airway geometries, resulting in the loss of information related to smaller branches. This study aims to introduce an unsupervised airway-mesh-smoothing learning (AMSL) method that preserves the original geometry of the three-dimensional (3D) airway for accurate CT-image-based CFD simulations. MethodThe AMSL method jointly trains two graph convolutional neural networks (GCNNs) defined on airway meshes to filter vertex positions and face normal vectors. In addition, it regularizes a combination of loss functions such as reproducibility, smoothness and consistency of vertex positions, and normal vectors. The AMSL adopts the concept of a deep mesh prior model, and it determines the self-similarity for mesh restoration without using a large dataset for training. Images of the airways of 20 subjects were smoothed by the AMSL method, and among them, the data of two subjects were used for the CFD simulations to assess the effect of airway smoothing on flow properties. ResultsIn 18 of 20 benchmark problems, the proposed smoothing method delivered better results compared with the conventional or state-of-the-art deep learning methods. Unlike the traditional smoothing, the AMSL successfully constructed 20 smoothed airways with airway diameters that were consistent with the original CT images. Besides, CFD simulations with the airways obtained by the AMSL method showed much smaller pressure drop and wall shear stress than the results obtained by the traditional method. ConclusionsThe airway model constructed by the AMSL method reproduces branch diameters accurately without any shrinkage, especially in the case of smaller airways. The accurate estimation of airway geometry using a smoothing method is critical for estimating flow properties in CFD simulations.