Identification Of Biomarkers In Gastric Adenocarcinoma With Heterogeneous Data Sources

Gastric adenocarcinoma(GA)is the common malignancy worldwide,with a relatively poor prognosis and a serious threat to human health.The occurrence of GA is disguised and indiscoverable in the early stage.The pathogenesis of GA is complex,and its specific etiology and pathogenesis need further study.Biomarkers with high specificity and sensitivity are of great significance for the diagnosis,targeted therapy,new drug development and prognosis analysis.Correlational studies have suggested that the abnormally expressed biomolecules goes hand in hand with the occurrence or development of diseases.In the process of molecular regulation,the complexity of pathways involved in each biomolecule is different.Also,each biomolecule has different effects on the stability and variability of physiological processes.Thus,those abnormally expressed molecules which play a key role in the upstream and downstream regulatory processes are more likely to be the biomarkers.The main research contents of this paper are as follows:(1)This work developed an algorithm for identifying potential mRNA biomarkers(Pm Bs)from complete transcriptomic profiles of gastric adenocarcinoma(GA)(called mRBioM).mRBioM firstly extracts differentially expressed(DE)RNAs(mRNA,micro RNA and lnc RNA).Then mRBioM calculates the amount of information expressed(AIE)by each DE mRNA based on the co-expression regulation relationships of three kinds of RNAs,and the amount of interaction information(AII)of each mRNA at protein level based on protein-protein interaction(PPI)network encoded by DE mRNAs.Next,the contribution rate of each DE mRNA to the occurrence and development of GA was calculated by combining the AIE and AII of each DE mRNA.Finally,the Pm Bs are identified according to the variation trend of DE mRNA contribution rates.mRBioM identified 55 Pm Bs(41 up-regulated and 14down-regulated)related to the occurrence and development of GA.Among them,more than 70% of Pm Bs have been revealed to be associated with a variety of cancers,thirteen Pm Bs have been verified to be biomarkers or potential therapeutic targets of gastric cancer and most of the Pm Bs are mainly enriched in the pathways closely related to the occurrence and development of gastric cancer.(2)Combining with the clinical information of patients,we used univariate and multivariate COX regression analysis to further screen out the survival-related Pm Bs and construct the prognostic risk assessment model.The result of COX regression analysis showed that LMNB2,BGN,MFSD12 and SOX4 selected from 55 Pm Bs were correlated with the prognosis of GA.The risk model constructed by these 4 Pm Bs could correctly classify 269 GA patients into the high-risk group(n=134)and the low-risk group(n=135),and the grouping results had significant statistical significance(***p < 0.001).(3)For confirming the reliability of identified Pm Bs,the cancer-related factor(CF)of a sample was determined by the expression values of 55 Pm Bs in the samples and was used to discriminate the sample types,which achieved the sensitivity,specificity and accuracy of 0.90,1 and 0.90 respectively.Furthermore,the normalized expression values of 55 Pm Bs were used as the features to design three sample classifiers: random forest(RF),support vector machine(SVM)and naive bayes(NB),which average accuracy,sensitivity and specificity of identifying GA-related samples ranged from0.94-0.98,0.94-0.97 and 0.97-1 respectively.(4)In order to confirm the generalization ability of mRBioM,we applied mRBioM to the complete transcriptomic profiles of colon adenocarcinoma,lung adenocarcinoma and liver hepatocellular carcinoma for identifying the Pm Bs.The three disease sample identification models constructed with Pm Bs all achieved GA equivalent classification performances.(5)In order to further explore the molecular mechanism related to the occurrence and development of GA and discover the ceRNAs that may affect the expression of Pm Bs,we started from Pm Bs and differentially expressed lnc RNAs,using public databases to predict the mi RNAs that target these two RNAs respectively.Using these mi RNAs as intermediary factors,we constructed the lnc RNA-mi RNA-mRNA ceRNA network that may exist as competing RNA pairs,and evaluated and analyzed the nature of the network.The ceRNA network contains 8 mRNAs,21 lnc RNAs,17 mi RNAs and63 lnc RNA-mi RNA-mRNA co-expression relationship pairs,which has typical biological network characteristics and has certain robustness.