Identification Of Cancer MiRNA/lncRNA Biomarkers Based On Multivariate Biomolecular Networks

With the proposal and advancement of precision medicine,screening non-coding RNAs as cancer biomarkers is now becoming a hot research topic.Among them,micro RNAs(miRNAs)and long non-coding RNAs(IncRNAs)are functional during cancer evolution.In particular,IncRNAs can act as competing endogenous RNAs(ceRNAs)to interplay with miRNAs,thereby regulating the expression of target genes.In the era of big data and biomedical informatics,computer-aided biomarker discovery has gradually become an emerging research paradigm.Based on multivariate biomolecular network analysis,it would be propitious to the systems-level identification and decoding of driving/key signals in the occurrence and development of complex diseases such as cancers,thus promoting the precision diagnosis and personalized therapy of diseases.In this study,we first reviewed the research progress in the field of computer-aided biomarker discovery,including data resources,computational models and applications.The result showed that most bioinformatics models lacked the guidance of theoretical rules Due to the high heterogeneity of sample data,the traditional 'training-testing' procedure used in some machine learning models often got overfitting results.To address these issues,we integrated 'miRNA-target' association data from different sources,and constructed miRNA-mRNA binary and IncRNA-miRNA-mRNA ternary networks to characterize the structural and functional traits of previously reported cancer miRNA/IncRNA biomarkers.The initiation and progression of disease could be recognized as the transition from one biological state to another,biomarkers hold the power to indicate the dynamical changes of biological states compared with disease-associated factors.Based on this idea,we focused on vulnerable sites in the network,and built systems-guided models to identify key miRNAs/lncRNAs as candidate biomarkers for cancer management.According to our research interests,in this study we selected prostate cancer metastasis as the case.Currently,metastasis is still the leading cause that affects the prognosis and survival of cancer patients.Therefore,it is of significance to screen miRNA/lncRNA biomarkers for prostate cancer metastasis.Based on network sub-structural analysis,we found that a number of mRNAs could be independently regulated by specific miRNAs in miRNA-mRNA network.We defined such regulatory mode as the single-line regulation.According to statistical tests,biomarker miRNAs had significantly stronger single-line regulation on mRNAs.Compared with multiple-line regulation,the single-line structure is special and vulnerable in the network,since the deregulation in these sites could not be substituted or compensated,and may cause the systems-level dysfunction,leading to the changes in biological states.Hence this structural pattern could be used as the theoretical basis for miRNA biomarker discovery.Based on this observation,we uncovered that biomarker miRNAs were also powerful to regulate/single-line regulate transcription factor genes.Thus,we integrated miRNA/mRNA expression profile data to extract miRNA-mRNA sub-network specific to prostate cancer metastasis,and optimized the bioinformatics model to identify miRNA biomarkers for predicting prostate cancer metastasis.The result showed that a total of five miRNAs,miR-204-5p,miR-101-3p,miR-145-5p,miR-198 and miR-152,could be used as potential biomarkers.Then we extended the above method into miRNA-mediated ternary network,and deciphered the regulation or competition features of miRNAs and lncRNAs in lncRNA-miRNA-mRNA network based on ceRNA hypothesis.The result indicated that compared with other miRNAs,miRNAs involved in IncRNA and mRNA competition could regulate more transcription factor genes,essential genes,house-keeping genes,and tumor-associated genes.In the network,several mRNAs and lncRNAs were single-line regulated by certain miRNAs,and a few of lncRNAs could bind more miRNAs,Based on the statistical evidences,we built a novel computational model to screen single and combinatory miRNAs/lncRNAs as candidate biomarkers for prostate cancer metastasis.In prostate cancer metastasis specific lncRNA-miRNA-mRNA sub-network,a total of 12 miRNAs,miR-23b-3p,miR-204-5p,miR-26b-5p,miR-27b-3p,miR-145-5p,miR-29b-3p,miR-143-3p,miR-130a-3p,miR-363-3p,miR-218-5p,miR-30c-5p,miR-101-3p,and five top-ranked IncRNAs,XIST,CTA-204B4.6,HCG18,TUG1,MALAT1,tended to have significantly higher regulatory power as single biomarker candidates.Moreover,considering the biological functions of mRNAs,these molecules could form as IncRNA-miRNA-mRNA combinations to indicate the states of prostate cancer invasion and metastasis.The bioinformatics validation showed that the identified candidates played important roles in prostate cancer associated pathways,such as prostate cancer,TFG-?signaling,etc.The result of qRT-PCR experiments using prostate cancer cell lines demonstrated that the expression of the candidates was significantly differed between primary prostate cancer cell line 22RV1 and metastatic cell lines LNCaP,PC3,or DU 145 Based on the comparison analysis,we found that the predictive results from different versions of the model were highly consistent.In particular,the three shared miRNAs,miR-204-5p,miR-145-5p,and miR-101-3p,were all significantly down-regulated in metastatic prostate cancer cell lines,which reflected their functional importance during prostate cancer progression.By contrast,the novel model has higher prediction precision and more powerful functions,which is reasonable and reliable for biomarker discoveryBased on multivariate biomolecular network analysis,in this study we proposed the principles for cancer miRNA/lncRNA biomarker discovery from a combinative perspective of network structures and functions,and constructed systems biology models to screen candidate miRNA/lncRNA biomarkers for cancers.The case of translational applications confirmed the performance of the models on identifying key miRNA/lncRNA signatures associated with prostate cancer metastasis.This study has both theoretical and clinical significance and the models could be applied to infer miRNA/lncRNA biomarkers for other cancers and complex diseases.