| Identifying key genes associated with carcinogenesis and progression from a large number of candidate genes plays a pivotal role in early screening and later treatment of diseases.With the development of unicellular sequencing technology,the huge amount of unicellular data provides an unprecedented opportunity for us to understand the pathogenesis of complex diseases.Therefore,it is of great importance to establish a statistical model that can integrate the data from different sources.Among them,gene network has realized the effective combination of network structure information and biological data information,which provides a more profound understanding of the pathogenesis of complex diseases.First of all,this dissertation analyzes the difference between the disease network and normal network.Based on DNA methylation data,gene co-expression networks in normal and disease states are constructed,and the differential reconnection information between network topologies is analyzed.Markov random field(MRF)based prediction method for key genes of cancer is proposed.Based on the MRF model,we model the prior network with the weight of the reconnection coefficient of the difference network.When the gene association label is a hidden variable,the single cell expression data follow the Gaussian mixture distribution.Finally,the key genes of cancer were predicted by gene prioritization and tested biologically.The AUC value of the MRF-based gene network model was 0.920,compared with the key gene prediction methods based on different network structural characteristics.The proposed method can fully integrate multimodal data and network information,improve the prediction accuracy of key cancer genes,and provide a basis for further cancer treatment and biomarker mining. |
