| Accurate identification of potential drug-target interactions is critical for drug development and repurposing.Although traditional biological experiment methods have achieved good results,they have the defects of high throughput,low precision,and high cost.The experimental verification of many drug-target interactions is blind and cannot be widely carried out in practical applications.Information processing technologies such as machine learning and data mining are flourishing and developing rapidly under the impetus of information science and technology,attracting the attention of researchers to computational methods aiming to identify drug-target interactions precisely through computational methods to guide biological experiments.Computational methods for drug-target interaction prediction can be divided into molecular docking and machine learning-based methods.Due to the shortcomings of a single method,it has become a development trend to identify drug-target interactions by proposing a combination method.This paper proposes a combined prediction algorithm integrating network and deep learning—GSRF-DTI.The main work and innovations of this paper are as follows:The GSRF-DTI model considers the influence of heterogeneous information and integrates multiple drug-associated and target-associated networks.Specifically,the isomorphic drug network is constructed by integrating the chemical structure network of the drug,the drug-drug interaction network,the drug-side effect association network,and the drugdisease association network;the protein sequence information network,the protein-protein interaction network,and the Protein-disease association network construct the target isomorphic network.Based on the constructed drug and target isomorphic network,the Deepwalk algorithm is used to learn the feature representation of drugs and targets,and finally,the 100-dimensional embedded features are retained.The GSRF-DTI model also considers the influence of the relationship between drug-target pairs on drug-target interaction prediction and uses a large-scale network representation learning algorithm to complete the feature learning of network nodes.Specifically,based on the drug set and target set,a drug-target pair network is constructed.The network nodes are composed of a drug and a target,and the corresponding drug and target features are stitched together as the initial features of the drug-target pair node(200-dimensional).For the edge of the network,it is stipulated that if two nodes share the same drug or the same target,it is considered that there is an edge connection between the two nodes.Otherwise,there is no edge.Based on the constructed drug-target pair network,the GraphSAGE algorithm is used to sample and aggregate neighbor information,and a 32-dimensional low-dimensional,dense,real-valued feature representation is obtained for subsequent drug-target interaction prediction.Based on the constructed drug-target pair network,the link prediction problem between nodes in the network is transformed into a binary classification problem of nodes.Based on the features obtained by the GraphSAGE algorithm,the random forest classifier is used to complete the drug-target interaction prediction.According to previous work experience,with AUROC and AUPR as evaluation indicators,5-fold cross-validation was used to evaluate the model’s effectiveness on Luo’s dataset.The experimental results obtained AUROC and AUPR values of 0.9818 and 0.9839,respectively,which are significantly better than some existing state-ofthe-art forecasting methods.The drug-target interaction prediction model proposed in this paper,which integrates multi-information networks and deep learning algorithms,has shown good performance and is expected to be transferred to similar biomedical problems,such as drug-drug interaction prediction,micro-RNA-disease interaction prediction,etc.At the same time,we should pay more attention to the practical applicability of the algorithm,continuously improve and design new algorithms,and genuinely contribute to the development of drugs and drug repositioning. |
