Grasping the intrinsic properties of protein structure is crucial for comprehending relevant biological mechanisms, with protein flexibility standing out as a critical aspect. Therefore, the prediction of protein flexibility is of great importance in understanding molecular mechanisms. We propose a deep learning method named ResaPred, which extracts diverse features from protein sequences, such as secondary structure, torsion angle, solvent accessibility, etc. ResaPred is a novel deep network based on a modified 1D residual module and a self-attention mechanism, which effectively extracts deep key features related to flexibility. The modified 1D residual module consists of three convolution layers, with batchnorm and relu layers added after each layer to prevent gradient explosion or vanishing. Incorporating self-attention mechanisms into neural network architectures introduces a significant advantage in capturing long-range dependencies within sequential data. We conduct experiments on the non-strict and strict cases, and achieve state-of-the-art results in predicting flexibility compared to existing methods. Furthermore, we extended our analysis to explore the correlation between protein secondary structure and solvent accessibility with flexibility. Finally, we used two important viral proteins as case studies, confirming the effectiveness of our method in recognizing the flexibility of protein structures.

The interaction between non-coding RNA (ncRNA) and protein (ncRPI) plays a crucial role in many physiological activities and disease progression. To identify ncRPIs on a large scale by computational methods based on deep learning is a common practice. However, existing computational methods face challenges such as low feature expression and redundant suppression capability when processing high dimensional feature data. To this end, we propose a new prediction model, called FSKansformer, in which a feature salience module is introduced to highlight useful information and suppress noise of multi-view feature matrices. In order to reduce the loss of feature information caused by serial extraction of global feature and local feature, we propose an parallel extraction framework in which a improved Kansformer is designed to extract global high-dimensional features, BiLSTM and LSTM techniques are used to extract local high-dimensional features simultaneously. Finally, the fused global-local high-dimensional features are input into the three-layer KAN network for dimensionality reduction to generate the final prediction score. Experiment results show that FSKansformer achieves state-of-the-art performance on five benchmark datasets compared with other models.

Combination therapy has exhibited substantial potential compared to monotherapy. However, due to the explosive growth in the number of cancer drugs, the screening of synergistic drug combinations has become both expensive and time-consuming. Synergistic drug combinations refer to the concurrent use of two or more drugs to enhance treatment efficacy. Currently, numerous computational methods have been developed to predict the synergistic effects of anticancer drugs. However, there has been insufficient exploration of how to mine drug and cell line data at different granularity levels for predicting synergistic anticancer drug combinations. Therefore, this study proposes a granularity-level information fusion strategy based on the hypergraph transformer, named HypertranSynergy, to predict synergistic effects of anticancer drugs. HypertranSynergy introduces synergistic connections between cancer cell lines and drug combinations using hypergraph. Then, the Coarse-grained Information Extraction (CIE) module merges the hypergraph with a transformer for node embeddings. In the CIE module, Contranorm is a normalization layer that mitigates over-smoothing, while Gaussian noise addresses local information gaps. Additionally, the Fine-grained Information Extraction (FIE) module assesses fine-grained information's impact on predictions by employing similarity-aware matrices from drug/cell line features. Both CIE and FIE modules are integrated into HypertranSynergy. In addition, HypertranSynergy achieved the AUC of 0.93${\pm }$0.01 and the AUPR of 0.69${\pm }$0.02 in 5-fold cross-validation of classification task, and the RMSE of 13.77${\pm }$0.07 and the PCC of 0.81${\pm }$0.02 in 5-fold cross-validation of regression task. These results are better than most of the state-of-the-art models.

RNA-binding proteins (RBPs) can regulate biological functions by interacting with specific RNAs, and play an important role in many life activities. Therefore, the rapid identification of RNA-protein binding sites is crucial for functional annotation and site-directed mutagenesis. In this work, a new parallel network that integrates the multi-head attention mechanism and the expectation pooling is proposed, named MAHyNet. The left-branch network of MAHyNet hybrids convolutional neural networks (CNNs) and gated recurrent neural network (GRU) to extract the features of one-hot. The right-branch network is a two-layer CNN network to analyze physicochemical properties of RNA base. Specifically, the multi-head attention mechanism is a computational collection of multiple independent layers of attention, which can extract feature information from multiple dimensions. The expectation pooling combines probabilistic thinking with global pooling. This approach helps to reduce model parameters and enhance the model performance. The combination of CNN and GRU enables further extraction of high-level features in sequences. In addition, the study shows that appropriate hyperparameters have a positive impact on the model performance. Physicochemical properties can be used to supplement characterization information to improving model performance. The experimental results show that MAHyNet has better performance than other models.

Accurately identifying potential drug-target interactions (DTIs) is a critical step in accelerating drug discovery. Despite many studies that have been conducted over the past decades, detecting DTIs remains a highly challenging and complicated process. Therefore, we propose a novel method called SMGCN, which combines multiple similarity and multiple kernel fusion based on Graph Convolutional Network (GCN) to predict DTIs. In order to capture the features of the network structure and fully explore direct or indirect relationships between nodes, we propose the method of multiple similarity, which combines similarity fusion matrices with Random Walk with Restart (RWR) and cosine similarity. Then, we use GCN to extract multi-layer low-dimensional embedding features. Unlike traditional GCN methods, we incorporate Multiple Kernel Learning (MKL). Finally, we use the Dual Laplace Regularized Least Squares method to predict novel DTIs through combinatorial kernels in drug and target spaces. We conduct experiments on a golden standard dataset, and demonstrate the effectiveness of our proposed model in predicting DTIs through showing significant improvements in Area Under the Curve (AUC) and Area Under the Precision-Recall Curve (AUPR). In addition, our model can also discover some new DTIs, which can be verified by the KEGG BRITE Database and relevant literature.

The intermolecular interactions between proteins and ligands occur through site-specific amino acid residues in the proteins, and the identification of these key residues plays a critical role in both interpreting protein function and facilitating drug design based on virtual screening. In general, the information about the ligands-binding residues on proteins is unknown, and the detection of the binding residues by the biological wet experiments is time consuming. Therefore, many computational methods have been developed to identify the protein-ligand binding residues in recent years. We propose GraphPLBR, a framework based on Graph Convolutional Neural (GCN) networks, to predict protein-ligand binding residues (PLBR). The proteins are represented as a graph with residues as nodes through 3D protein structure data, such that the PLBR prediction task is transformed into a graph node classification task. A deep graph convolutional network is applied to extract information from higher-order neighbors, and initial residue connection with identity mapping is applied to cope with the over-smoothing problem caused by increasing the number of graph convolutional layers. To the best of our knowledge, this is a more unique and innovative perspective that utilizes the idea of graph node classification for protein-ligand binding residues prediction. By comparing with some state-of-the-art methods, our method performs better on several metrics.

Transcription factors are a class of protein factors that bind directly or indirectly to RNA polymerases and regulate the initiation of transcription by recognizing cis-acting elements in the DNA sequence. The prediction of transcription factor binding sites is an important part of the study of gene transcriptional regulation. Therefore, accurate prediction of TFBS helps one to understand and study the spatiotemporal nature of transcriptional regulation of target genes by different transcription factors. In recent years, an increasing number of deep learning methods have been used to predict transcription factor binding sites, however, existing methods still much room to improve performance. In this paper, we present a deep learning framework combining convolutional neural networks and recurrent neural networks to predict transcription factor binding sites, called DeepGenBind, for the systematic identification of transcription factor binding sites from DNA sequences. The novelty of our proposed approach relies on two key aspects: (1) the framework combines a three-layer parallel convolutional neural network CNN with a two-layer LSTM to efficiently extract useful features from large-scale genomic sequences obtained by high-throughput sequencing techniques (2) the use of k-mer coding to transform DNA sequences, with the transformed short sequences allowing for better data reading. Experimental results on 165 datasets from ENCODE show that DeepGenBind outperforms several other state-of-the-art methods in identifying transcription factor binding sites. In addition, we tested the effect of varying the k-mer vector length on model performance, demonstrating the variation in model performance under different k-mer related parameter settings. Overall, DeepGenBind is a useful tool for the cost-effective and accurate identification of potential transcription factor binding sites in biological genomes.

In the drug discovery process and repurposing of existing drugs, accurately identifying ligands with high binding affinity to proteins is a very critical step. However, it sinks a lot of time and resources to detect the protein-ligand binding affinity through biological experiments. Therefore, it is very necessary to develop an accurate and reliable computational method to predict the binding affinity between protein and ligand. At present, some computational methods have been proposed to predict the protein-ligand binding affinity, but the absence of protein-ligand complexes structures restricts some predictive methods that require input the complexes structures. In this paper, a novel deep-learning-based method is proposed, named DPLA, to predict binding affinity by integrating multilevel information of protein and ligand. More specifically, our model extracted some important information, such as sequence representation, structural property representation of amino acids in protein and protein binding pocket, MACCS key ligand molecular fingerprint and ligand molecular network features. This method was tested on the PDBbind core set, and we compared it with some recent state-of-art protein-ligand affinity prediction methods. The excellent performance shows that DPLA is an accurate and reliable method for affinity prediction.

BACKGROUND: The study of DNA binding protein (DBP)-drug interactions can open a breakthrough for the treatment of genetic diseases and cancers. Currently, network-based methods are widely used for protein-drug interaction prediction, and many hidden relationships can be found through network analysis. We proposed a DCA (drug-cluster association) model for predicting DBP-drug interactions. The clusters are some similarities in the drug-binding site trimmers with their physicochemical properties. First, DBPs-drug binding sites are extracted from scPDB database. Second, each binding site is represented as a trimer which is obtained by sliding the window in the binding sites. Third, the trimers are clustered based on the physicochemical properties. Fourth, we build the network by generating the interaction matrix for representing the DCA network. Fifth, three link prediction methods are detected in the network. Finally, the common neighbor (CN) method is selected to predict drug-cluster associations in the DBP-drug network model. RESULT: This network shows that drugs tend to bind to positively charged sites and the binding process is more likely to occur inside the DBPs. The results of the link prediction indicate that the CN method has better prediction performance than the PA and JA methods. The DBP-drug network prediction model is generated by using the CN method which predicted more accurately drug-trimer interactions and DBP-drug interactions. Such as, we found that Erythromycin (ERY) can establish an interaction relationship with HTH-type transcriptional repressor, which is fitted well with silico DBP-drug prediction. CONCLUSION: The drug and protein bindings are local events. The binding of the drug-DBPs binding site represents this local binding event, which helps to understand the mechanism of DBP-drug interactions.

The first joint fingerprinting and decryption (JFD) for vector quantization (VQ) images addressed the problem that the decrypted multimedia data may be redistributed from authorized customers to unauthorized customers. The scheme also caused conventional JFD methods to be equipped with a special ability to resist noise interference. Till now, some existing schemes related have been proposed to protect the multimedia content and distribution, but these schemes failed to tackle several problems existing in the original JFD scheme based on VQ image, including high transmission cost and severe fingerprinted image distortion. In this paper, we propose a novel JFD method by combining a weight-sum function with fingerprinting embedding and extraction for VQ images. Under the combination, the visual quality of the fingerprinted image is further improved; also the fingerprint extraction implements a blind extraction process. Experiments and analyses demonstrate the feasibility of the proposed method.

BACKGROUND: DNA-binding proteins perform important functions in a great number of biological activities. DNA-binding proteins can interact with ssDNA (single-stranded DNA) or dsDNA (double-stranded DNA), and DNA-binding proteins can be categorized as single-stranded DNA-binding proteins (SSBs) and double-stranded DNA-binding proteins (DSBs). The identification of DNA-binding proteins from amino acid sequences can help to annotate protein functions and understand the binding specificity. In this study, we systematically consider a variety of schemes to represent protein sequences: OAAC (overall amino acid composition) features, dipeptide compositions, PSSM (position-specific scoring matrix profiles) and split amino acid composition (SAA), and then we adopt SVM (support vector machine) and RF (random forest) classification model to distinguish SSBs from DSBs. RESULTS: Our results suggest that some sequence features can significantly differentiate DSBs and SSBs. Evaluated by 10 fold cross-validation on the benchmark datasets, our prediction method can achieve the accuracy of 88.7% and AUC (area under the curve) of 0.919. Moreover, our method has good performance in independent testing. CONCLUSIONS: Using various sequence-derived features, a novel method is proposed to distinguish DSBs and SSBs accurately. The method also explores novel features, which could be helpful to discover the binding specificity of DNA-binding proteins.