With the growing application of deep learning in drug design, efficiently generating molecules with desirable drug-like properties and target specificity remains a key challenge. Existing methods often rely on complex fusion of molecular and protein features, leading to high data demands and training costs. To address this, a novel molecule generation model based on multi-constraint optimization is proposed, incorporating docking scores, Lipinski’s Rule of Five, and other pharmacological constraints to guide generation. The model is built on a Bidirectional Gated Recurrent Unit (BiGRU) and improves training efficiency through cross-entropy loss optimization. Experimental results show that the proposed approach achieves superior performance in molecular validity, novelty, and uniqueness, and generates compounds with lower binding energies to target proteins, significantly enhancing the practicality and scalability of molecular design.

Single-cell sequencing technology has enabled researchers to study cellular heterogeneity at the cell level. To facilitate the downstream analysis, clustering single-cell data into subgroups is essential. However, the high dimensionality, sparsity, and dropout events of the data make the clustering challenging. Currently, many deep learning methods have been proposed. Nevertheless, they either fail to fully utilize pairwise distances information between similar cells, or do not adequately capture their feature correlations. They cannot also effectively handle high-dimensional sparse data. Therefore, they are not suitable for high-fidelity clustering, leading to difficulties in analyzing the clear cell types required for downstream analysis. The proposed scSAMAC method integrates contrastive learning and negative binomial losses into a variational autoencoder, extracting features via contrastive unit similarity while preserving the intrinsic characteristics. This enhances the robustness and generalization during the clustering. In the contrastive learning, it constructs a mask module by adopting a negative sample generation method with gene feature saliency adjustment, which selects features more influential in the clustering phase and simulates data missing events. Additionally, it develops a novel loss, which consists of a soft k-means loss, a Wasserstein distance, and a contrastive loss. This fully utilizes data information and improves clustering performance. Furthermore, a multi-head attention mechanism module is applied to the latent variables at each layer of autoencoder to enhance feature correlation, integration, and information repair. Experimental results demonstrate that scSAMAC outperforms several state-of-the-art clustering methods.

This paper describes a targeted drug design method based on a framework of multiscale encoder-decoder. Encoders are used to encode target gene and protein features. A decoder is used to design drugs based on target features. This method fuses target gene and protein information for targeted drug design, and invokes effective feature extraction strategies. A multilevel gene feature extraction (MGFE) is proposed to extract multilevel target features by extracting base and codon features in gene expression. The process of extracting features from nucleotide sequences by MGFE based gene encoder simulates the process of gene transcription and translation. Meanwhile, a multi-embedding protein feature extraction (MPFE) is proposed to extract target protein features from amino acid sequences. The MPFE based protein encoder includes three embedding layers which provides a unique linear layer for each amino acid. According to structural characteristics of proteins, amino acids with different positions but the same type are embedded into the same embedding vector without location encoding. Finally, a gated recurrent unit based drug decoder is used to decode gene and protein features, and creates new targeted drugs. The experiments adminstrate that the proposed method outperforms the previous ones in terms of validity, novelty and binding affinity.

Accurate drug dosage in cancer treatments depends on precise liver and tumor size estimation, but tumors pose challenges for automatic segmentation due to their small, scattered volumes, complex structures, and low contrast against surrounding organs. Conventional U-Net models struggle to capture these microstructural details. To address this, we propose LKC-Net, a liver and tumor segmentation method leveraging large kernel parallel dilated convolutions. The encoder utilizes depth-wise separable large kernels combined with parallel dilated convolutions and squeeze-and-excitation (SE) modules. This architecture enhances the receptive field and improves the capture of small-scale patterns. SE modules filter redundant information, focusing the model on key regions. Experiments on ATLAS, LiTS, and private datasets demonstrate LKC-Net’s superior performance in liver and tumor segmentation tasks.

Drug-target interaction (DTI) prediction has an important role in drug discovery, significantly expediting the drug design and development process. This paper proposes a new model, ERW_BiAN, that seamlessly integrates graph embedding representations with deep learning network. Specifically, we utilize an improved graph embedding model, ERW, to generate comprehensive feature vectors for each node within the knowledge graph. Subsequently, these feature vectors are fed into the BiAN model, which leverages an attention mechanism to assign precise weights to sequences. Experimental results demonstrate the superior accuracy and predictive prowess of ERW_BiAN in DTI prediction tasks. Furthermore, the case about COVID-19 shows that ERW_BiAN has the better application potential for predicting drug-target interactions.

Accurate detection of protein binding sites is critical for facilitating drug design. Most existing models rely on features at a single scale, leading to the loss of crucial global or local structural information. Therefore, this paper proposes a novel 3D model (ResPocket), to fully capture the multi-scale features of proteins. First, Residual Block (ResB) based on U-Net is used to capture features from different scales, including global and local structural information. Then, an Enhanced Attention Module (EAM) is designed to encode semantic dependencies by effectively capturing global context information. In addition, ResPocket focuses on learning local information about protein pockets with Self-Guided Mask Block (SGMB), to increase the robustness of the overall model and reduce its dependence on external mask guidance. Experimental results show that ResPocket improves by an average of 4.89% on DCA Top-n, by an average of 1.48% and 2.25% on DCC and DVO metrics, respectively.

This paper describes a multi target drug design method based on features of the target proteins. The multi target drug which inhibit multiple proteins have prospective applications, but design is difficult. In this study, the target protein sequences are embedded to obtain target features. The features of each target are independently encoded as target latent vectors, and the features of multiple targets are jointly encoded as target similarity latent vectors. Based on the features of targets and the similarity features among the targets, multi target drugs are efficiently designed. In the experiments, the designed multi target drugs can be docked with target proteins with a proprietary molecular structure according to the requirements of different target pocket structures. The excellent fitness among the molecular structure of multi target drugs and the protein structure of multiple targets confirms the performance of the proposed method. In the molecular docking part of the experi¬ments, the binding affinity of designed multi target drugs is far better than that in the previous studies, which administrates the better performance of this work.

Accurate prediction of drug-drug interactions (DDIs) plays an important role in improving the efficiency of drug development and ensuring the safety of combination therapy. Most existing models rely on a single source of information to predict DDIs, and few models can perform tasks on biomedical knowledge graphs. This paper proposes a new hybrid method, namely Knowledge Graph Representation Learning and Feature Fusion (KGRLFF), to fully exploit the information from the biomedical knowledge graph and molecular structure of drugs to better predict DDIs. KGRLFF first uses a Bidirectional Random Walk sampling method based on the PageRank algorithm (BRWP) to obtain higher-order neighborhood information of drugs in the knowledge graph, including neighboring nodes, semantic relations, and higher-order information associated with triple facts. Then, an embedded representation learning model named Knowledge Graph-based Cyclic Recursive Aggregation (KGCRA) is used to learn the embedded representations of drugs by recursively propagating and aggregating messages with drugs as both the source and destination. In addition, the model learns the molecular structures of the drugs to obtain the structured features. Finally, a Feature Representation Fusion Strategy (FRFS) was developed to integrate embedded representations and structured feature representations. Experimental results showed that KGRLFF is feasible for predicting potential DDIs.

Computer-aided drug design and artificial intelligence-driven drug design have accelerated drug discovery. However, how to design effective drugs that have strong interaction ability with target proteins to further improve the efficacy of drugs in treating diseases remains a key issue. This paper proposes a new target-specific drug generative model 3CLpro2mol to generate new drug molecules, which uses features of drug-target interactions (DTIs) to constrain the correlation between the drug and the target protein. To obtain as many drug-target interaction features as possible from a small amount of data, a small molecule extraction strategy is proposed to ensure the diversity of small molecules in the training samples. To improve the efficiency and accuracy of the generative model, a TOP K sampling strategy is used to generate tokens, which can improve the rationality and diversity of the generated molecules. The experimental results show that the proposed model has the potential to generate small molecules that interact better with the target protein.

Most current representations of protein pockets are the atom-pair graph, which ignore the global structural information of amino acids. Therefore, we propose a new molecular generation model, which uses the hypergraph to represent protein pocket structure, and combines the structural features obtained by atom-pair graph representation. These two levels of graphs are more capable of representing the complex structural information of protein pockets. Then, the graphs of the two levels of the protein pockets are input into the improved network model, which is named Hypergraph Graph Attention Fusion (HGAF), to obtain the embedding representation of the protein pockets, which is used as a condition to constrain the molecule generation. The molecules sampled by HGAF are subjected to quality assessment and docking targeting validation. Experimental results show that the molecules generated by proposed method can achieve better results in both of these assessment approaches.

Existing works on document image shadow removal mostly depend on learning and leveraging a constant background (the color of the paper) from the image. However, the constant background is less representative and frequently ignores other background colors, such as the printed colors, resulting in distorted results. In this paper, we present a color-aware background extraction network (CBENet) for extracting a spatially varying background image that accurately depicts the background colors of the document. Furthermore, we propose a background-guided document images shadow removal network (BGShadowNet) using the predicted spatially varying background as auxiliary information, which consists of two stages. At Stage I, a background-constrained decoder is designed to promote a coarse result. Then, the coarse result is refined with a background-based attention module (BAModule) to maintain a consistent appearance and a detail improvement module (DEModule) to enhance the texture details at Stage II. Experiments on two benchmark datasets qualitatively and quantitatively validate the superiority of the proposed approach over state-of-the-arts.

It is challenging to precisely segment the liver from surrounding organs in medical images because of the poor contrast between them. A method of semantic segmentation of 3D liver images based on multi-path features attention mechanism is proposed to address this issue. It integrates three-dimensional spatial information and feature information from several paths in the model to automatically segment the liver area. The model in this paper uses the LiTS dataset for training, testing, and ablation experiments, and compares the results with previous models. The experimental results demonstrate that the model in this paper has reached 0.965 in the DICE similarity coefficient, and has also improved in evaluation indicators such as volume overlap error (VOE) and root mean square symmetric surface distance (RMSD). It also has better segmentation performance when tested on the CHAOS dataset. Cross-validation was carried out on the clinical MRI dataset of a hospital, and the DICE similarity coefficient reached 0.971. The results show that the model has good performance on the multi-modal datasets of CT and MRI.

The emergence of single-cell RNA-seq (scRNA-seq) technology makes it possible to capture their differences at the cellular level, which contributes to studying cell heterogeneity. By extracting, amplifying and sequencing the genome at the individual cell level, scRNA-seq can be used to identify unknown or rare cell types as well as genes differentially expressed in specific cell types under different conditions using clustering for downstream analysis of scRNA-seq. Many clustering algorithms have been developed with much progress. However, scRNA-seq often appears with characteristics of high dimensions, sparsity and even the case of dropout events', which make the performance of scRNA-seq data clustering unsatisfactory. To circumvent the problem, a new deep learning framework, termed variational graph attention auto-encoder (VGAAE), is constructed for scRNA-seq data clustering. In the proposed VGAAE, a multi-head attention mechanism is introduced to learn more robust low-dimensional representations for the original scRNA-seq data and then self-supervised learning is also recommended to refine the clusters, whose number can be automatically determined using Jaccard index. Experiments have been conducted on different datasets and results show that VGAAE outperforms some other state-of-the-art clustering methods.

Computational prediction of drug-target interaction (DTI) is important for the new drug discovery. Currently, the deep neural network (DNN) has been widely used in DTI prediction. However, parameters of the DNN could be insufficiently trained and features of the data could be insufficiently utilized, because the DTI data is limited and its dimension is very high. To deal with the above problems, in this paper, a graph auto-encoder and multi-subspace deep neural network (GAEMSDNN) is designed. GAEMSDNN enhances its learning ability with a graph auto-encoder, a subspace layer and an ensemble layer. The graph auto-encoder can preserve the reconstruction information. The subspace layer can obtain different strong feature subsets. The ensemble layer in the GAEMSDNN can comprehensively utilize these strong feature subsets in a unified optimization framework. As a result, more features can be extracted from the network input and the DNN network can be better trained. In experiments, the results of GAEMSDNN are significantly improved compared to the previous methods, which validates the effectiveness of our strategies.

Residual image and illumination estimation have been proven to be helpful for image enhancement. In this article, we propose a general framework, called RI-GAN, that exploits residual and illumination using generative adversarial networks (GANs). The proposed framework detects and removes shadows in a coarse-to-fine fashion. At the coarse stage, we employ three generators to produce a coarse shadow-removal result, a residual image, and an inverse illumination map. We also incorporate two indirect shadow-removal images via the residual image and the inverse illumination map. With the residual image, the illumination map, and the two indirect shadow-removal images as auxiliary information, the refinement stage estimates a shadow mask to identify shadow regions in the image, and then refines the coarse shadow-removal result to the fine shadow-free image. We introduce a cross-encoding module to the refinement generator, in which the use of feature-crossing can provide additional details to promote the shadow mask and the high-quality shadow-removal result. In addition, we apply data augmentation to the discriminator to reduce the dependence between representations of the discriminator and the quality of the predicted image. Experiments for shadow detection and shadow removal demonstrate that our method outperforms state-of-the-art methods. Furthermore, RI-GAN exhibits good performance in terms of image dehazing, rain removal, and highlight removal, demonstrating the effectiveness and flexibility of the proposed framework.

Effective identification of potential drug-drug interactions (DDIs) can prevent adverse effects caused by DDIs to a certain extent. This paper proposes the new hybrid method for predicting DDIs, which is called BiRW-KGBAN that combines bidirectional random walk and back aggregation based on knowledge graph. BiRW-KGBAN first constructs two knowledge graph KG-DrugBank and KG-KEGG that integrate data related to drugs. Then, an improved sampling method based on bidirectional random walk (BiRW) is used to sample the neighborhood information of drugs, including directly connected entities, related semantic relations and potential information. In addition, the path with drug as the center node is also extracted. Finally, two improved back aggregation model KGBAN1 and KGBAN2 are used to obtain the final embedded representation of the drugs. The experiments show that, compared with other existing methods, our method could obtain the higher-order topological information and the deeper potential neighborhood information of drugs, and has great improvements in DDIs prediction. The case study also shows that the proposed method has the potential for actual DDIs prediction.

BACKGROUND: Computational prediction of the interaction between drugs and protein targets is very important for the new drug discovery, as the experimental determination of drug-target interaction (DTI) is expensive and time-consuming. However, different protein targets are with very different numbers of interactions. Specifically, most interactions focus on only a few targets. As a result, targets with larger numbers of interactions could own enough positive samples for predicting their interactions but the positive samples for targets with smaller numbers of interactions could be not enough. Only using a classification strategy may not be able to deal with the above two cases at the same time. To overcome the above problem, in this paper, a drug-target interaction prediction method based on multiple classification strategies (MCSDTI) is proposed. In MCSDTI, targets are firstly divided into two parts according to the number of interactions of the targets, where one part contains targets with smaller numbers of interactions (TWSNI) and another part contains targets with larger numbers of interactions (TWLNI). And then different classification strategies are respectively designed for TWSNI and TWLNI to predict the interaction. Furthermore, TWSNI and TWLNI are evaluated independently, which can overcome the problem that result could be mainly determined by targets with large numbers of interactions when all targets are evaluated together. RESULTS: We propose a new drug-target interaction (MCSDTI) prediction method, which uses multiple classification strategies. MCSDTI is tested on five DTI datasets, such as nuclear receptors (NR), ion channels (IC), G protein coupled receptors (GPCR), enzymes (E), and drug bank (DB). Experiments show that the AUCs of our method are respectively 3.31%, 1.27%, 2.02%, 2.02% and 1.04% higher than that of the second best methods on NR, IC, GPCR and E for TWLNI; And AUCs of our method are respectively 1.00%, 3.20% and 2.70% higher than the second best methods on NR, IC, and E for TWSNI. CONCLUSION: MCSDTI is a competitive method compared to the previous methods for all target parts on most datasets, which administrates that different classification strategies for different target parts is an effective way to improve the effectiveness of DTI prediction.

Cover Caption: The cover image is based on the Research Article Efficient virtual data search for annotationfree vehicle reidentification by Zhijing Wan et al., https://doi.org/10.1002/int.22829.

Vehicle reidentification (re-ID) is the task of retrieving the same vehicle across nonoverlapping cameras, which has made significant progress with the help of abundant manually annotated real images. To avoid the time-consuming and tedious labeling of real images, virtual data sets with large-scale synthetic images have recently been constructed to perform annotation-free model training. However, current methods fail to exploit the potential of virtual data search, that is, searching valuable and representative virtual subdata set for efficient training. This paper presents a novel data sampling strategy from both semantic and feature levels to perform an effective data search. The semantic level determines the sample number of each vehicle identity via the consistency constraint of attribute distribution for source domain and target domain; while the feature level searches valuable and representative samples of each vehicle identity. To our knowledge, we are among the first attempts to search effective virtual data to perform annotation-free vehicle re-ID. Extensive cross-domain experiments from virtual vehicle re-ID data sets to real vehicle re-ID data sets show that our data sampling strategy can significantly reduce the training data volume and even boost the re-ID performance.

Lymph-node metastasis is the most perilous cancer progressive state, where long non-coding RNA (lncRNA) has been confirmed to be an important genetic indicator in cancer prediction. However, lncRNA expression profile is often characterized of large features and small samples, it is urgent to establish an efficient judgment to deal with such high dimensional lncRNA data, which will aid in clinical targeted treatment. Thus, in this study, a local linear reconstruction guided distance metric learning is put forward to handle lncRNA data for determination of cancer lymph-node metastasis. In the original locally linear embedding (LLE) approach, any point can be approximately linearly reconstructed using its nearest neighborhood points, from which a novel distance metric can be learned by satisfying both nonnegative and sum-to-one constraints on the reconstruction weights. Taking the defined distance metric and lncRNA data supervised information into account, a local margin model will be deduced to find a low dimensional subspace for lncRNA signature extraction. At last, a classifier is constructed to predict cancer lymph-node metastasis, where the learned distance metric is also adopted. Several experiments on lncRNA data sets have been carried out, and experimental results show the performance of the proposed method by making comparisons with some other related dimensionality reduction methods and the classical classifier models.

This paper proposes a NPEL (Nine-Pipeline & Ensemble Learning) strategy based on machine Learning algorithm to predict protein-protein interaction hotspots by training amino acid composition, surface area, amino acid chains and other complex/interface-related structural information. We applied Random Forest, Linear Svm, KNN, Gaussian Naive Bayes, Multi-layer Perceptron Neural Network, Adaboost, XGBoost etc. nine machine learning algorithms combination into an independent pipeline to predict protein hot spots, and the final results are optimized through voting and stacking scheme. In the stacking result of XGBoost and Logistic Regression, the highest accuracy is 0.8462 and improve the indicators of the pipeline results greatly.

generating new molecules is very important for drug design. Currently, many deep generative models have been designed, such as variational autoencoder (VAE), adversarial autoencoder (AAE), and reinforcement learning (RL). However, many problems are also existed in these models. Firstly, the information among molecules could be not utilized in optimizing these models. Secondly, some useful molecule information could be not used, such as fingerprint. Thirdly, the information contained in different molecule representations could be not used together. To overcome the above problems, in this paper, a multi-label learning and adversarial autoencoder (MLAAE) based de novo drug design method is designed. MLAAE enhances its learning ability with a new designed multi-label classifier and a new designed double AAEs collaborative optimization framework. These new designs can learn a better latent space whose global distribution is similar with the random distribution, but local distribution contains much information. As a result, the generator can be trained by more information, and the input of generator in testing is similar with that in training. The conducted experiments validate the effectiveness of our MLAAE.

Drug discovery is of great significance in medical and biological research, while the study of Drug-Target Interaction (DTI) and Drug-Drug Interaction (DDI) can help accelerate drug discovery progress. This paper proposes a new hybrid method for DTI prediction and DDI prediction, which is called MHRW2Vec-TBAN that combines graph representation learning and neural network. MHRW2VecTBAN first constructs knowledge graph KG-DTI and KG-DDI that integrate data related to drugs and targets. Then, an improved graph representation learning model, MHRW2Vec model, is used to obtain feature vectors of reflecting the network structure information for improving the performance of representation learning. Finally, the feature vectors obtained are input to the improved neural network model TextCNN-BiLSTM-Attention Network (TBAN). The experimental results show that, compared with other existing methods, our method could discover more deeper the relationship between drugs and their potential neighborhoods, and has great improvements in DTI prediction and DDI prediction. In addition, the case study of prediction COVID-19 DTI also shows that the proposed model has the potential for actual drug discovery.

Aiming at the problems of difficult segmentation, time-consuming and low precision of 3D liver medical images, 3D liver image semantic segmentation method based on context attention strategy is proposed. This method combines the three-dimensional spatial boundary information on the upper liver features map and the overall channel information on the lower liver features map. The model used the LiTS data set for ablation experiments in comparison with the previous model. Experimental results show that the Dice similarity coefficient (DICE) is increased by 1.6%, and the volume overlap error (VOE) is reduced by 2.49% in comparison with Channel-Unet. The relative volume difference (RVD), the average symmetric surface distance (ASD) and the root mean square symmetric surface distance (RMSD) which indicates that the algorithm improves the performance of 3D liver medical images segmentation. Extended experiments were carried out on the Sliver07 data set, the CT data set in CHAOS and the clinical MRI liver medical imaging data set of a hospital, and the results indicated that the proposed method has strong generalization ability.

BACKGROUND: In the process of designing drugs and proteins, it is crucial to recognize hot regions in protein-protein interactions. Each hot region of protein-protein interaction is composed of at least three hot spots, which play an important role in binding. However, it takes time and labor force to identify hot spots through biological experiments. If predictive models based on machine learning methods can be trained, the drug design process can be effectively accelerated. RESULTS: The results show that different machine learning algorithms perform similarly, as evaluating using the F-measure. The main differences between these methods are recall and precision. Since the key attribute of hot regions is that they are packed tightly, we used the cluster algorithm to predict hot regions. By combining Gaussian Naïve Bayes and DBSCAN, the F-measure of hot region prediction can reach 0.809. CONCLUSIONS: In this paper, different machine learning models such as Gaussian Naïve Bayes, SVM, Xgboost, Random Forest, and Artificial Neural Network are used to predict hot spots. The experiment results show that the combination of hot spot classification algorithm with higher recall rate and clustering algorithm with higher precision can effectively improve the accuracy of hot region prediction.

A multigraph is modeled as a bag of graphs. Exact multigraph matching search aims to find all multigraphs that are the same as the query multigraphs from the data multigraph datasets. To the best of our knowledge, works regarding exact multigraph matching search have not been reported although they have a very wide range of application scenarios. In this article, we propose an efficient algorithm to solve the problem of exact multigraph matching search. We first propose a definition of exact multigraph matching and its Basic Method (BM), called BM, which has a considerable amount of graph isomorphism detection calculations and, thus, has very high computational complexity. Obviously, it is impractical to compare the query multigraph to each data multigraph in the multigraph datasets. To reduce the search space, multiple filtering conditions are proposed to obtain a candidate result set containing all the final results, including the cardinality filter, the vertex filter, the edge filter, the size filter, and the star filter. Then, each multigraph in the candidate result set is verified with the Improved BM (IBM) algorithm. Moreover, an offline and Multilayer Inverted Index (MII), named MII, is proposed to further accelerate the search process. Finally, we propose an Exact Multigraph Matching Search (EMMS) algorithm, based on the abovementioned technologies. We also analyze its time complexity. Extensive experiments on real datasets demonstrate the effectiveness and efficiency of the proposed algorithms.

Low light images captured in a non-uniform illumination environment usually are degraded with the scene depth and the corresponding environment lights. This degradation results in severe object information loss in the degraded image modality, which makes the salient object detection more challenging due to low contrast property and artificial light influence. However, existing salient object detection models are developed based on the assumption that the images are captured under a sufficient brightness environment, which is impractical in real-world scenarios. In this work, we propose an image enhancement approach to facilitate the salient object detection in low light images. The proposed model directly embeds the physical lighting model into the deep neural network to describe the degradation of low light images, in which the environment light is treated as a point-wise variate and changes with local content. Moreover, a Non-Local-Block Layer is utilized to capture the difference of local content of an object against its local neighborhood favoring regions. To quantitative evaluation, we construct a low light Images dataset with pixel-level human-labeled ground-truth annotations and report promising results on four public datasets and our benchmark dataset.

Residual images and illumination estimation have been proved very helpful in image enhancement. In this paper, we propose a general and novel framework RIS-GAN which explores residual and illumination with Generative Adversarial Networks for shadow removal. Combined with the coarse shadow-removal image, the estimated negative residual images and inverse illumination maps can be used to generate indirect shadow-removal images to refine the coarse shadow-removal result to the fine shadow-free image in a coarse-to-fine fashion. Three discriminators are designed to distinguish whether the predicted negative residual images, shadow-removal images, and the inverse illumination maps are real or fake jointly compared with the corresponding ground-truth information. To our best knowledge, we are the first one to explore residual and illumination for shadow removal. We evaluate our proposed method on two benchmark datasets, i.e., SRD and ISTD, and the extensive experiments demonstrate that our proposed method achieves the superior performance to state-of-the-arts, although we have no particular shadow-aware components designed in our generators.

Computer-aided drug discovery can efficiently predict drug-target interactions, which helps biological researchers to narrow down the search space and reduce experimental consumption. However, the accuracy of existing prediction methods still needs to be further improved. This paper proposes a deep learning model that predicts drug-target interactions through effective strategies: the graphical representation of SMILES based on Graph Isomorphic Network (GIN) and word vector matrices of amino acid sequences based on one-dimensional convolution, which is used to extract features of drug-target pairs for classification prediction. Compared with the previous methods, the proposed method outperforms the existing work with AUC. A case study of predicting COVID-19 DTIs also shows that the proposed method can be invoked to be use for practical drug-target prediction.

Computational prediction of drug-target interaction (DTI) is very important for the new drug discovery. However, by connecting drugs and targets to form drug target pairs, the number of interactions is limit, most interactions focus on only a few targets or a few drugs, and the number of drug target pairs is far more than the number of interactions, which causes to be over fitting. To overcome the above problem, in this paper, a multiple output deep neural network (MODNN) based DTI prediction is designed. MODNN enhances its learning ability with a kind of auxiliary classifier layers. The parameters used in the training process are elaborated from the auxiliary and main classifier layers, which can increase the gradient signal that gets propagated back, utilize multi-level features to train the model, and use the features produced by the higher, middle or lower layers in a unified framework. The conducted experiments validate the effectiveness of our MODNN.

Abstract In this paper, we propose a novel context and lightness aware Generative Adversarial Network (CLA‐GAN) framework for shadow removal, which refines a coarse result to a final shadow removal result in a coarse‐to‐fine fashion. At the refinement stage, we first obtain a lightness map using an encoder‐decoder structure. With the lightness map and the coarse result as the inputs, the following encoder‐decoder tries to refine the final result. Specifically, different from current methods restricted pixel‐based features from shadow images, we embed a context‐aware module into the refinement stage, which exploits patch‐based features. The embedded module transfers features from non‐shadow regions to shadow regions to ensure the consistency in appearance in the recovered shadow‐free images. Since we consider pathces, the module can additionally enhance the spatial association and continuity around neighboring pixels. To make the model pay more attention to shadow regions during training, we use dynamic weights in the loss function. Moreover, we augment the inputs of the discriminator by rotating images in different degrees and use rotation adversarial loss during training, which can make the discriminator more stable and robust. Extensive experiments demonstrate the validity of the components in our CLA‐GAN framework. Quantitative evaluation on different shadow datasets clearly shows the advantages of our CLA‐GAN over the state‐of‐the‐art methods.

Proteins are not isolated biological molecules, which have the specific three-dimensional structures and interact with other proteins to perform functions. A small number of residues (hot spots) in protein-protein interactions (PPIs) play the vital role in bioinformatics to influence and control of biological processes. This paper uses the boosting algorithm and gradient boosting algorithm based on two feature selection strategies to classify hot spots with three common datasets and two hub protein datasets. First, the correlation-based feature selection is used to remove the highly related features for improving accuracy of prediction. Then, the recursive feature elimination based on support vector machine (SVM-RFE) is adopted to select the optimal feature subset to improve the training performance. Finally, boosting and gradient boosting (G-boosting) methods are invoked to generate classification results. Gradient boosting is capable of obtaining an excellent model by reducing the loss function in the gradient direction to avoid overfitting. Five datasets from different protein databases are used to verify our models in the experiments. Experimental results show that our proposed classification models have the competitive performance compared with existing classification methods.

Discovering hot regions in protein-protein interaction is important for understanding the interactions between proteins, while because of the complexity and time-consuming of experimental methods, the computational prediction method can be very helpful to improve the efficiency to predict hot regions. In previous researches, some models are based on a single aspect, such as structure, energy, and sequence, each aspect has its advantage and limitations. In this paper, a new method that combing structure-based classification, energy-based clustering and sequence-based conservation in evolution is proposed. This method makes full use of three aspects of protein information and compensates for the limitations of using one single aspect information. Experimental results show that the proposed method significantly improves the prediction performance of hot regions.

Hot spot residues bring into play the vital function in bioinformatics to find new medications such as drug design. However, current datasets are predominately composed of non-hot spots with merely a tiny percentage of hot spots. Conventional hot spots prediction methods may face great challenges towards the problem of imbalance training samples. This paper presents a classification method combining with random forest classification and oversampling strategy to improve the training performance. A strategy with an oversampling ability is used to generate hot spots data to balance the given training set. Random forest classification is then invoked to generate a set of forest trees for this oversampled training set. The final prediction performance can be computed recursively after the oversampling and training process. This proposed method is capable of randomly selecting features and constructing a robust random forest to avoid overfitting the training set. Experimental results from three data sets indicate that the performance of hot spots prediction has been significantly improved compared with existing classification methods.

The hot regions in PPIs are some assembly regions which are composed of the tightly packed HotSpots. The discovery of hot regions helps to understand life activities and has very important value for biological applications. The identification of hot regions is the basis for protein design and cancer prevention. The existing algorithms of predicting hot regions often have some defects, such as low accuracy and unstability. This paper proposes a novel hot region prediction method based on diverse biological characteristics. First, feature evaluation is employed by using an impoved mRMR method. Then, SVM is adopted to create cassification model based on the features selected. In addition, a new clustering algorithm, namely LCSD (Local community structure detecting), is developed to detect and analyze the conformation of hot regions. In the clustering process, the link similarity of protein residues is introduced to handle the boundary nodes. This algorithm can effectively deal with the missing residue nodes and control the local community boundaries. The results indicate that the spatial structure of hot regions can be obtained more effectively, and that our method is more effective than previous methods for precise identification of hot regions.

Stacked auto-encoder is mainly used for image classification and it can extract valid information from data through unsupervised pre-training and supervised fine-tuning. This paper is intended to improve the accuracy of image classification, we constructed a 6-layer stacked convolution neural network (CNN) based on stacked auto-encoders. The constructed CNN can extract effective features for image classification through greedy layer-wise training. In order to make the constructed CNN to have strong robustness to noise, we added a noisy-layer in the pre-training stage. Adding the sparsity constraint can make the training of the CNN more effective, and can also reduce data redundancy. For classification applications, our experiments show that the final classification results of the proposed model is superior to the combination of auto-encoders and the noisy auto-encoders.

In the protein-protein interactions, only a small subset of hot spot residues contributes significantly to the binding free energy. Therefore, there is an imbalance between the number of hot spots and non-hot spots. The prediction of hot spot residues is very important in the protein-protein interaction. This paper presents an improved ensemble learning method-Adaboost with SMOTE method to deal with the imbalanced data and predict protein hot spots in the latest database SKEMPI. Firstly, the amino acid information such as hydrophobicity of the amino acid and protein structural features is exacted. Then mRMR algorithm was used to select the features. Finally, the protein database is further handled by SMOTE to deal with the imbalance data, the protein hot spots are predicted by the ensemble learning method-Adaboost. Experimental results show that the proposed method has the ability to improve the predict accuracy.

Proteins play a crucial role in every organism, which perform a vast amount of functions. The hot regions in protein-protein interactions consist of hot spot residues in protein-protein binding sites which are called interfaces, can help proteins to perform their biological function. Residue based computational prediction of hot regions might be useful to understand the molecular mechanism and is crucial in drug design and protein design. However, it is very challenging to identify the hot regions in protein-proteins. In this paper, we have proposed a support vector machine based on ensemble learning system for predicting hot spot residues, and predicted hot regions in protein-protein interactions. The efficiency of our method is analyzed in identifying hot spots and hot regions in protein-protein interactions and the results obtained are compared with the existing techniques. The results demonstrate that the proposed method is superior to identify the hot spots and hot regions in the protein interfaces.

Dimensionality reduction is indispensable for high dimensional data classification. So in this paper, a novel supervised method is developed to reduce dimensions of the original data, which is named feature space distance metric learning (FSDML). Instead of distances between any two points, distances between any two feature spaces are involved in the proposed method. Besides feature space distances(FSD) metric, the inter-class data separablity and the intra-class data locality are all employed for graph embedding, by which a subspace will be explored for data discriminant analysis. The proposed FSDML are evaluated by some state-of-art methods such as linear discriminant analysis (LDA), unsupervised discriminant projection (UDP) and nearest feature space embedding (NFSE). Experiments on some benchmark data sets including AR and ORL face data have shown that the proposed method is effective and efficient.

Saliency computational model with active environment perception can substantially facilitate a wide range of applications. Conventional saliency computational models primarily rely on hand-crafted low level image features, such as color or contrast. However, they may face great challenges in low lighting scenario, due to the lack of well-defined feature to represent saliency information in low contrast images. In this paper, we propose a novel deep neural network framework embedded with covariance descriptor for salient object detection in low contrast images. Several low-level features are extracted to compute their mutual covariance, which is then trained via a 7-layers convolutional neural network (CNN). The saliency map can be generated by estimating the final saliency score of each region via the pre-trained CNN model. Extensive experiments have been conducted on six challenging datasets to evaluate the performance of the proposed model against ten state-of-the-art models.

This paper studies how to select input parameters in density clustering of hot region prediction. There are two parameters radius and density in density-based incremental clustering. We firstly fix density and enumerate radius to find a pair of parameters which leads to maximum number of clusters, and then we fix radius and enumerate density to find another pair of parameters which leads to maximum number of clusters. Experiment results show that the proposed method using both two pairs of parameters provides better prediction performance than the other method, and compare these two predictive results, the result by fixing radius and enumerating density have slightly higher prediction accuracy than that by fixing density and enumerating radius.

We examined the conservation of hot regions in protein-protein interaction in evolution for the first time. We studied sequence conservation of hot regions in different species annotated in the database ASEdb, applying the BLOSUM matrix to construct conservation scoring function of hot regions. The experimental results showed that there is high correlation of conservation between hot regions in different species. We further tested the conservation of hot regions in different species annotated in the latest SKEMPI database using the same method. Our experimental results suggest that there is obvious conservation in hot regions.

Predicting the spacial folding structure of a protein, given its sequence of amino acids, is one of the central problems in computational biology field. This paper studies the AB off-lattice model with two species of monomers, called hydrophobic (A) and hydrophilic (B). Based on this simplified model, the low energy configurations are searched by using the GAPSO. A kind of optimization about the mutation mechanism and the Euclidean interference mechanism are presented, where a novel local adjustment strategy is also used to enhance the searching ability of the global minimum within the AB off-lattice model. Starting from random conformations, the GAPSO method can find the low-energy conformation of the Fibonacci sequences and the real protein sequences. Compared with other optimization methods, the proposed novel method could converge to the lower energy folds. It appears that the proposed method can used for solving protein folding problem, which is based on the thermodynamic hypothesis.

There is an important relationship between the stability of protein complex and hot region. Research has shown that in protein-protein interaction (PPI), residues are denser around the hot region. Therefore, this paper proposed an algorithm based on Gi statistics, regional division rule and regional amplification principle to form residue dense region (RDR); Then, according to the results of cascade classifier composed of Naive Bayes and Back-Propagation (BP) neural network classifier, non-hotspot residues in RDRs were removed; At length, we used binding free energy change value calculated from Robetta Server to modify predicted hot regions. The experimental results showd that the proposed method can effectively improve the prediction accuracy on hot regions.

Accompanying increasing competition among the communication industry, maintaining and improving the stability and loyalty of customers has become the key determinant of profitability. In order to prevent the loss of customers, we need to identify the stable users by data mining model. Through the evaluation of three models, Random Forest model performs with better robustness. This model can describe and predict most of the stable users in a shorter period of time. Consequently, the result will provide operators with the advantage of adopting reasonable marketing tactics timely.

BACKGROUND: Protein folding structure prediction is one of the most challenging problems in the bioinformatics domain. Because of the complexity of the realistic protein structure, the simplified structure model and the computational method should be adopted in the research. The AB off-lattice model is one of the simplification models, which only considers two classes of amino acids, hydrophobic (A) residues and hydrophilic (B) residues. RESULTS: The main work of this paper is to discuss how to optimize the lowest energy configurations in 2D off-lattice model and 3D off-lattice model by using Fibonacci sequences and real protein sequences. In order to avoid falling into local minimum and faster convergence to the global minimum, we introduce a novel method (SATS) to the protein structure problem, which combines simulated annealing algorithm and tabu search algorithm. Various strategies, such as the new encoding strategy, the adaptive neighborhood generation strategy and the local adjustment strategy, are adopted successfully for high-speed searching the optimal conformation corresponds to the lowest energy of the protein sequences. Experimental results show that some of the results obtained by the improved SATS are better than those reported in previous literatures, and we can sure that the lowest energy folding state for short Fibonacci sequences have been found. CONCLUSIONS: Although the off-lattice models is not very realistic, they can reflect some important characteristics of the realistic protein. It can be found that 3D off-lattice model is more like native folding structure of the realistic protein than 2D off-lattice model. In addition, compared with some previous researches, the proposed hybrid algorithm can more effectively and more quickly search the spatial folding structure of a protein chain.

In the protein-protein interactions (PPI), hot regions are the key factor to maintain metabolism and the cause of disease. The conformation and prediction of the hot regions effectively is a worth researching topics. This paper proposes a prediction method of hot regions based on complex network and community detection. In the prediction process, a hotspots retrieving strategy is used to exploit false positive (FP) and false negative (FN) residues, and these FP and FN residues achieved from the previous hotspots prediction. The method has advantage to identify hot regions in PPI, and experimental results show that the method not only improves the prediction accuracy of hot regions, but also has more reliability.

Blind noisy image quality assessment aims to evaluate the quality of the degraded noisy image without the need for the ground truth image. To tackle this challenge, this paper proposes an image quality assessment approach using block homogeneity. The contribution of the proposed approach is two-fold. First, a block-based homogeneity measure is proposed to estimate the statistics (e.g., variance) of the noise incurred in the image, based on adaptively selected homogeneous image regions. Second, an image quality assessment approach is proposed by exploiting the above-mentioned estimated noise variance, along with the visual masking effect of the human visual system. Experimental results are provided to demonstrate that the proposed image noise estimation approach yields superior accuracy and stability performance to that of conventional approaches, and the proposed image quality assessment approach achieves consistent performance to that of human subjective evaluation.

Traditional classification algorithms are difficult in dealing with imbalance data. This paper proposes a classification algorithm called CascadeBoost, which combines with the advantages of boosting algorithm and cascade model that can learn imbalance data. Cascade model allows the pre-training data to be balanced by gradually reducing the number of the major class; and then the most rich information samples based on the weight distribution can be gradually selected using boosting algorithm. The experimental results show that the proposed method obtains better performance compared to other methods.

This paper presents a new structural feature for feature-based retinal image registration. The conventional point-matching methods largely depend on the branching angles of single bifurcation point. The feature correspondence across two images may not be unique due to the similar angle values. In view of this, structure-matching registration is favored. The bifurcation structure is composed of a master bifurcation point and its three connected neighbors. The characteristic vector of each bifurcation structure consists of the normalized branching angle and length, which is invariant against translation, rotation, scaling, and even modest distortion. This can greatly reduce the ill-posed nature of the matching process as long as the vasculature pattern can be segmented. The simplicity and efficiency of the proposed method make it readily to be applied alone or incorporated with other existing methods to formulate a hybrid or hierarchy scheme.

Web application has become one of the main network applications. The security in web application systems is very important. In this paper we present a method for tracking suspicious users based on web data analysis. Based on a browser/server model, internet flow data are stored in the server and outliers considered as the suspicious users are detected with web data analysis. The behaviors of these suspicious users are tracked and those browsed web pages are recorded in database. By replaying the visited web pages, the proposed method is able to monitor the given suspicious users, which can benefit intranet network security.

With the wide applications of video cameras in surveillance, video analysis technologies have attracted the attention from the researchers in computer vision field. In video analysis, human behavior recognition and understanding is an important research direction. By recognition and understanding the human behaviors, we can predict and recognize the happening of crimes and help to the police or other agencies to react immediately. In the past, large amount of intensive papers have been published on human behavior understanding in videos. Generally speaking, the procedure of human behavior understanding can be divided into the following stages: human segmentation and tracking, and human behavior recognition. In this paper, we provide a comprehensive survey of the recent development of all these stages. We will also discuss the difficulties in behavior understanding and identify possible future directions.

The widespread networks in campus provide students with rich resources, but they also provide unhealthy information which may have negative effects on students. Filtering systems have been developed to prevent web users from unhealthy web contents. However, the false positive and false negative rates of those tools are still high and thus need to be improved. In this paper, we propose an Internet behavior monitoring method by integrating the BHO plug-in techniques with URL-based filtering technique, and apply this method in the campus network. Designed in the client/server mode, the method has the capabilities of filtering out unhealthy websites according to the given black list, and can record the students' online web activities into XML documents and/or pictures. The recorded contents can be replayed by the campus network administrators. This method assists teachers to keep eyes on students' learning state, and notifies the administrators to react immediately to help campus web users correct their inappropriate behaviors.

Image enhancement technologies have been widely used for improving the quality of the images for screening mammograms. In this paper, we will focus on the enhancement of breast calcifications, which are the deposits of calcium that can be seen on a mammogram of the breast. In the proposed method, the original image and the normalized gradient image of the original image are first decomposed into a multi-level Laplacian pyramids, and then the features in different scales are enhanced level by level based on a contrast measure during the reconstruction stage. Because the importance of different levels is different, different weights are used in different levels. Experiments proved the effectiveness of the proposed algorithm.

Due to the biological immune system applied to the field of computer security, immunological scientists have made much development for anomaly detection systems. However, there are still a number of significant hurdles to prevent it from solving real-world problems efficiently, such as the high false positive and false negative errors. In order to present a more feasible anomaly detection system, we outline multi-agent systems (MAS) to design an artificial immune system inspired by a novel immune theory- danger theory, following an appropriate evaluation tool (DCs) for network packets and a suitable mechanism of communication between agents. We set up two kinds of immune responses logically on both host layer and network layer to the coming intruders for the purpose of mitigating the damage and infection. We hope that this system will eventually become more powerful as a distributed immune system, based on the sound immunological concepts.