Identification Of Cancer Diagnostic And Prognostic Biomarkers Based On Multi-omics Data

Cancer is one of the major causes of death in China.Cancer is generally characterized by difficulty in early diagnosis,poor prognosis and easy recurrence.At present,although many cancer treatments have been developed,the death rate of cancer is still high.Researches on mechanism of cancer is the key to find better treatments.In which,the identification of cancer-related biomarkers can better promote the development of cancer early screening and molecular targeted drugs.Therefore,this paper took cancer as the research object.Mult-omics data including mRNA expression,DNA methylation,copy number variation,somatic mutation,micro RNA expression and lncRNA expression were integrated to constract bioinfomatics method to identify biomarkers associated with diagnosis and prognosis.1)Abnormal methylation can cause genetic instability,which can further lead to cancer.Lnc RNA has also been shown to be closely associated with cancer.In this paper,mainly based on the methylation changes of lncRNA genes,methylation of lncRNA genes,lncRNA expression and mRNA expression data were integrated to identify the differentially methylated lncRNAs.Subsequently,they were screened at the levels of survival correlation,negative correlation between methylation changes and expression changes and ceRNA to obtain lncRNA diagnostic biomarkers.Finally,a total of 29 lncRNA diagnostic biomarkers for 10 kinds of cancers and two common lncRNA diagnostic markers HOXA-AS2 and AC007228 were obtained.Identification of lncRNA diagnostic markers related to methylation can provide more assurancefor the accuracy of cancer diagnosis.2)The occurrenceof cancer is caused by the disorder of multi-genes.In order to find the mode of multi-gene collaboration,this paper propose multi-DBnet method,integrated mRNA expression,copy number variation,somatic mutation,miRNA expression and lncRNA expression data to study genes in the form of network,and identified key gene modules in the network.For each gene,a regression model was constructed and adaptive Lasso was used for model training to identify disregulated genes between with and without mutation samples.Then the disregulatory genes were put into protein-protein interaction network,and the correlation between disregulatory genes were calculated to identify disregulatory gene modules.Finally,a ten gene module(APC,TSC2,CTNNB1,RB1,EPB41L3,AXIN1,CAND1,BAP1,GPR63 and CARD11)was identified as a biomarker for hepatocellular carcinoma.And 11 genes(PBRM1,RB1,NF1,OBSCN,TP53,CDKN2 A,PTEN,NIPBL,SMAD4,BAP1 and PARD3)were identified disorderd in more than four cancers.The combination of multi-gene diagnostic markers with immunohistochemistry and genetic testing can improve the accuracy of cancer identification and help precision medicine of cancer.3)The identification of cancer prognostic markers can help screen patients who need intensive monitoring or adjuvant therapy to intervene in adverse natural outcomes.In this paper,DNA methylation,mRNA expression,copy number variation and miRNA expression data were integrated.A method for identifying prognostic markers of cancer: multi-PB was proposed.Applying the method,ten genetic prognostic biomarkers were identified for 13 types of cancer,respectively.And genes related to survival of multi-cancer were found: SLK,ZRANB1,BTBD2,PTAR1,VPS37 A,EIF2B1 and API5.Then,comparing the results with other methods,the Cindex of multi-PB was obviously superior to other methods that also integrated multiomics data.Multi-PB can more accurately identify prognostic biomarkers,which can help predict the prognostic status of patients and intervene in patients with poor natural prognosis as early as possible.4)Prognosis and survival prediction of cancer patients can help to better select the appropriate treatment strategy for patients.In order to construct a more accurate prognosis and survival prediction model for patients,this paper integrated mRNA expression,DNA methylation,copy number variation,miRNA expression and lncRNA expression data.A prognostic prediction method: multi-SP was proposed.In this method,feature screening was performed by using a strategy by support vector machine based on recursive feature elimination.Then,the graph convolution network was used to train the classifier model,and the view association discovery network was used to integrate the classification results,so as to obtain the final prognostic classification of patients.During the classification process,the accuracy of the prediction by the classification of one,three and five years survival was assessed respectively.The results showed that the selected features in each group were closely related to hepatocellular carcinoma,and the prediction model was best when the 5-year classification standard was used(～0.967).Subsequently,the method was extended to renal clear cell carcinoma.Multi-SP uses deep learning to construct a prediction model,which can predict patient prognosis more accurately than other machine learning methods,especially 5-year survival.In summary,this work integrated multi-omics data for cancer research,not only identified the genes and modules related to the diagnosis and prognosis of cancer,but also constructed the survival prediction model.This work identified specific and common biomarkers for cancers.The biomarkers identified in this work can be used as candidate molecules for cancer screening and targeted drug trials.Compared with previous methods,the multi-DBnet,multi-PB and multi-SP methods proposed in this work can be applied to many types of cancer.The results of this paper can provide help for the diagnosis,treatment and pathogenesis of cancer.