Research On Prediction Algorithm Of Gene Regulatory Network Based On Attention Mechanism And Multi Source Information Fusion

Gene regulatory networks control the effects of gene expression throughout development and adulthood in plants and animals,and also play important roles in various cellular developmental processes and division pathways.Gene transcriptome data include single-cell transcriptome data and bulk transcriptome data,which contain rich intercellular gene regulation information.Through gene transcriptome data,the regulatory relationship of various genes in the expression process can be discovered.In this paper,two deeplearning models,DeepGRN and DeepTGI,are proposed to predict the interaction between transcription factors and target genes from gene transcriptome data.Construct gene regulatory networks and predict potential transcriptional regulatory relationships.The main work content of this paper is as follows:(1)For the training and testing of the model,six gene expression data of mESC,bulk,mHSC-E,mHSC-L,hESC and h Hep and the corresponding gene regulatory network datasets were collected,and the datasets were cleaned at the same time and preprocessing.(2)In order to predict gene regulatory network from transcriptome data,a gene regulatory network prediction algorithm(DeepGRN)based on Auto-Encoder and attention mechanism is proposed.The algorithm obtains attention scores through attention mechanism,uses Auto-Encoder for feature extraction,and uses artificial neural network to predict gene regulation network.Training and testing were performed on m HSC-E,m HSC-L,h ESC and h Hep,and the performance was compared with four popular gene regulatory network prediction algorithms on the test set.(3)In order to further improve the prediction performance of the gene regulatory network,an improved multi-source information fusion deeplearning algorithm(DeepTGI)is proposed on the basis of DeepGRN.The algorithm first uses methods such as Auto-Encoders,convolutional neural networks,and Siamese neural networks to extract multi-source features of transcriptome data and perform feature fusion.Attention score feature extraction is then performed on the fused features,followed by gene regulatory network prediction.Trained and tested on m ESC and bulk datasets,and compared performance with DeepGRN and seven other popular gene regulatory network prediction algorithms.The experimental results show that the two algorithms proposed in this paper,DeepGRN and DeepTGI,are superior to the existing algorithms in terms of the performance of transcription factor and target gene interaction prediction and gene regulatory network prediction,and the performance of the improved DeepTGI algorithm is also better than that of DeepGRN algorithm.