| The study of gene function mainly associated with gene sequence information with gene function through various omics,algorithms and biological experimental techniques.The study of gene sequence and function not only reveals the laws of life activities at different levels of biological systems in nature,but also is closely related to human disease prevention,new drug development and resistance genes,promoting the development of genetics,developmental biology,medicine,agriculture and other disciplines.In the information age,algorithm,as one of the important tools to solve problems,can quickly obtain the required output in a short time through the pre-processing information identified by the input computer,and has been widely used in various fields.In our study,the sequence and function of genes are studied by mathematical modeling.The first part of our study apply deep learning to HIV-1 virus on the chromatin area integration sites prediction of integration sites and the surrounding a sequence,k_mer after segmentation,carries on the word vector training,and then build a deep learning model,to join the attention mechanism,the greater the influence of features to give the greater the weight,improve the model performance(AUROC increased about 0.01).We construct different weak classifiers by random sampling of training sets and finally adopt model average strategy ensemble learning.In deep learning,we compared the traditional one_hot and the word vector model-based word2 vec,doc2vec and GloVe,and found that GloVe had the highest model performance,with AUROC at 0.881 and AUPRC at 0.879.Compared with other models,AUROC is superior to lstm-cnn model of Min et al.We balance the positive and negative samples,so the AUPRC is also 0.511 higher than that of Hailin et al.,and the AUROC is slightly higher than 0.879 of Hailin et al.Our model has higher performance and potential,and the effectiveness of the model is illustrated by parameter optimization.In the second part of our study,based on the recommendation algorithm,the interaction between lncrnas and proteins was predicted through the similarity network of multiple lncrnas and proteins.We each took 3000 lncRNA,protein,calculated the total similarity,gene expression of their sequence similarity,building a network,through the resumption of the characteristics of the random walk algorithm to learn lncRNA,protein,and finally by bilinear function mapping principle of learning space,have not found the lncRNA-protein interaction relationship between score predicts.After 10-fold cross-validation,the accuracy of the model was 0.971 and the AUROC reached 0.986,which was better than the 0.968 of the PLPIHS model of Xiao et al.The model can provide ideas for subsequent massive gene interactions,discover new lncrna-protein interactions,and study the functions of lncRNA. |
