| Data exists in nature in a variety of forms,expressing connections between things through structured and unstructured data,which is reflected in every aspect of daily life.How to reasonably use multi-source data and integrate data information is the next important research content in the era of big data.In the field of Drug discovery,highly efficient integration of multi-source data of drugs and targets and prediction of potential drug-target interactions(DTIs)can help shorten the time and improve the efficiency of drug development.Traditional DTIs prediction has problems of insufficient data and low prediction efficiency.Therefore,this thesis focuses on the efficient fusion of multi-source data and the design of DTIs prediction model,efficient fusion of multi-source data based on similarity network fusion(EFMS-SNF)was proposed.In addition,a DTIs prediction algorithm based on graph autoencoder and gradient boosting decision tree(GAEGBDT-DTIs)was designed,with specific studies as follows:(1)A multi-source data fusion algorithm(EFMS-SNF)based on similarity network fusion is proposed.Considering the topological structure information and semantic information characteristics of the data,the fusion strategies of multi-source data is designed,and a method of multi-source data fusion based on similarity network fusion(EFMS-SNF),is proposed to realize the fusion of multi-source network data into a network,so that the fusion network contains more potential information.The multi-source data of drugs and targets were classified according to the data sources,and the contribution of different networks in DTIs prediction was analyzed by combining the prediction experimental results of different drugs and targets.Four fusion strategies were designed to obtain the optimal fusion network.Experimental results show that EFMS-SNF algorithm has better performance in the prediction task(AUROC is 0.95,AUPR is0.96).Meanwhile,the first 18 prediction results are supported by the literature.(2)A DTIs prediction model based on graph autoencoder and decision tree(GAEGBDTDTIs)is proposed.Based on the multi-source data fusion network of drugs and targets,a network embedding algorithm based on the graph encoder was designed to obtain the low-dimensional feature vectors of drugs and targets,and the score of drug-target pair was predicted based on the decision tree.Compared with the four most advanced prediction models,GAEGBDT-DTIs showed significant improvement in AUROC and AUPR.The results show that multi-source data processing and feature extraction algorithms are particularly outstanding in the calculation of drug-target interaction.Therefore,considering the efficient fusion of multi-source data fully,the prediction model can be designed to predict DTIs more effectively. |
