Research On The Survival Prediction Model Of Pancreatic Cancer Based On Machine Learning

Pancreatic cancer has become the seventh factor of cancer death worldwide,with an increasing incidence and poor prognosis for patients.Determining individualized treatment strategies according to different prognostic criteria is the critical factor to enhancing the outcomes of patients.Based on the genetics and clinical data of pancreatic cancer patients,this paper established a prognostic model through machine learning methods,identified key genes and prognostic markers related to the happenings and evolution of this cancer,and enriched a theoretical foundation for enhancing the outcomes and prognosis of patients.The main elements of this artical are as follows:1.Research on the prediction model of overall survival of patients based on support vector regression and recursive feature elimination method.In this paper,a hybrid algorithm of support vector regression and recursive feature elimination was used to construct the first model for quantitative prediction of overall survival(OS)in patients with pancreatic cancer,and 70 RNAs related to OS were identified,including 33 m RNAs,28 lnc RNAs and 9 mi RNAs.The results of 10-fold cross-validation(R~2 is 0.9693)and the generalization ability(R~2 is 0.9666)showed that the model has reliable predictive performance and these 70 RNAs are important factors influencing the OS of pancreatic cancer patients.To further study the relationship between RNA-RNA interaction and the survival,competitive endogenous RNA regulation network was constructed.Degree centrality,betweenness centrality and closeness centrality of nodes in the ce RNA network showed that hsa-mir-570,hsa-mir-944,hsa-mir-6506,hsa-mir-3136,MMP16,PLGLB2,HPGD,FUT1,MFSD2A,SULT1E1,SLC13A5,ZNF488,F2RL2,TNFRSF8,TNFSF11,FHDC1,ISLR2 and THSD7B are hub nodes,which are key RNAs closely determining the OS of pancreatic cancer patients.2.Construction of a prognostic model for pancreatic cancer patients based on deep learning and Cox models.A deep neural network model was proposed to predict the mortality risk of pancreatic cancer patients.The model not only considered the nonlinear relationship between covariates and solved the problem that the traditional Cox model only considered the linear relationship of variables,but also taked into account gene co-expression relationships and the linear and nonlinear relationships between biological pathways.The model achieved better prediction results than existing models such as Deep Surv,and the C-index increased from0.7869 to 0.8023.Enrichment analysis showed that 61 co-expressed genes were significantly enriched in 27 biological processes including extracellular space,immune response modulation,digestion,skin development,lipid digestion,lipase activity,B cell activation cell surface and30 biological pathways including pancreatic secretion,primary immune deficiency,glycerolipid metabolism,amebiasis and metabolic pathway.Further studies have shown that these biological processes and pathways play a significant role in the happenings and evolution of pancreatic cancer,and further analysis of co-expressed genes and their enrichment results will help to improve the prognosis of pancreatic cancer patients.3.Identification of key molecules for pancreatic cancer prognosis based on PPI network.425 genes revelant to the prognosis of pancreatic cancer patients were screened from 3699differentially expressed genes through univariate Cox analysis,and then their corresponding protein-protein interaction network was constructed.Based on complex network theory,KIF20A,BUB1,CEP55,ASPM,CDK1,TOP2A,CENPF,TPX2,PBK,KIF2C,CCNA2,CCNB2,NUSAP1,CDC20,TTK,BUB1B,KIF11,KIF23,NCAPG and DLGAP5,etc.key nodes were identified,and these genes are key molecules associated with pancreatic cancer prognosis.A prognostic risk model was constructed through multivariate Cox regression analysis.Separated all patients into two groups according to their risk scores and the results of the survival analysis showed that there was an obviously diversity in the OS between them.Risk score can serve as an independent prognostic factor in pancreatic cancer patients.In this paper,the prognostic models of pancreatic cancer were constructed through machine learning.The characteristic molecules in the model can provide a reference for the individualized therapy methods of pancreatic cancer and provide ideas for the precise treatment of patients.