Research On Gene Regulatory Network Construction Algorithm Based On Single-Cell Multi-Omics Data

The gene regulatory network can provide us with more comprehensive insights into complex biological processes,and have been regarded as an effective way to clarify the complex molecular interactions in biological processes and systems.Most of the molecular interactions in the gene regulatory network are inferred from a single data type,such as gene expression data.However,the expression of a gene is the product of many biological processes,such as DNA sequence variation,copy number variation,histone modification,transcription factor and DNA methylation.Therefore,integrating multi-omics data can provide more information for the construction of gene regulatory networks.However,some methods are mainly applied to the multi-omics data obtained by Bulk sequencing,and it is difficult to conduct a more detailed analysis.The emergence of single-cell multi-omics data provides more opportunities for researchers to construct gene regulatory networks from the single-cell level.At the same time,machine learning methods are suitable for large data,which can more easily find rules from known data and make predictions.This is helpful for integrating single-cell multi-omics data to build gene regulatory networks.For a long time,building gene regulatory networks with single cell resolution has been an enormous challenge.Therefore,this paper proposes two algorithms to integrate single-cell multi-omics data to construct gene regulatory networks.The main work of this paper is as follows:(1)An algorithm integrating single-cell RNA sequencing data and DNA methylation data to construct a gene regulatory network based on the back-propagation(BP)neural network is proposed(sc BPGRN).This algorithm uses biweight extreme correlation coefficients to measure the correlation between factors and uses neural networks to calculate generalized weights to construct gene regulation networks.Finally,the node strength is calculated to identify the genes associated with cancer.We apply the sc BPGRN algorithm to hepatocellular carcinoma(HCC)data.we construct a regulatory network and identify top-ranked genes,such as MYCBP,KLHL35,PRKCZ,and SERPINA6,as the key HCC-related genes.In addition,the single-cell data is found to consist of two subpopulations.We also apply sc BPGRN to two subpopulations.The consequences of functional enrichment analysis indicate that the gene regulatory network we have constructed is valid.Our results have been verified in several pieces of literature.This study provides a reference for the integration of single-cell multi-omics data to construct gene regulatory networks.(2)A new method is proposed to joint graph convolution neural network and bayesian network to construct gene regulatory network and classify cells(sc MOGRN).Firstly,the graph convolution neural network is used to integrate the protein-protein interaction network and copy number variation data.The link probability between two nodes can be calculated according to the gene characteristics.The unknown regulatory relationship will be predicted based on the known regulatory relationship,and they are used as the biological prior information of the Bayesian network.Then the Bayesian network is used to integrate gene expression data,and the Markov Chain Monte Carlo method is used to solve the posterior probability to obtain the optimal network structure.Finally,the score of each gene is calculated according to the node sorting algorithm.The cells are classified according to the top genes.Our algorithm is compared with other algorithms on simulated datasets.The results show that our algorithm is better.In addition,our algorithm is also applied to the colorectal cancer dataset to classify 134 single cells.The results are consistent with the data source.These results show that the gene regulatory network constructed by sc MOGRN in this paper is useful and can provide assistance for cell classification.