Drug Repositioning Prediction And Cancer Consensus Molecular Subtype Identification Based On Omics Big Data

With the rapid development of technologies such as gene sequencing,high-throughput gene chips and low-cost transcriptomes,a large number of multi-omics data(genome,transcriptome,proteome,metabolome,etc.)about drug perturbation,gene knockdown trials and cancer have been accumulated.These massive data not only profoundly affect the basic biology research,but also provide substantial opportunities for the development and applications of drug repositioning and cancer consensus molecular subtype identification.In addition,machine learning,especially deep learning,has been developed rapidly in recent years,and deep learning has been successfully applied in computer vision,speech recognition and natural language processing.The combination of drug big data and deep learning will provide new development opportunities in biomedical fields,including drug R&D.Entire research is divided into two parts: the first part is drug repositioning prediction based on transcriptional data and deep learning,and the second part is a web server about cancer consensus molecular subtypes identification based on multi-omics data fusion.In the first part,we constructed gene co-expression networks induced by different therapeutic properties drugs based on drug-induced transcriptional profile of LINCS project.There is a core gene co-expression network(DGCN)by analyzing topological properties and biological function of these networks.The co-expression association of genes in DGCN can be used to explore drug-induced gene expression patterns,so as to help researchers shed light on the mode of action of drugs.Next,we constructed a deep neural network model(DNN)to systematically predict drug-target interaction based on transcriptional data from drug perturbation and gene knockout trials in the L1000 database of LINCS project.After training our model,the results suggested that the performance of the DNN was much better than other prediction algorithms by cross-validation,and predicted novel drug-target interactions were significantly enriched in other relevant databases.In the second part,we developed a web tool named COMSUC,based on multi-omics data of TCGA,ICGC and TARGET projects and several clustering methods.COMSUC made up for the inconsistency of molecular subtypes caused by different sequencing platforms,preprocessing processes and clustering methods.COMSUC implemented a variety of application scenarios and clustering algorithms,which can visualize,interactive and publishable output the analysis results.Moreover,we took the identification of consensus molecular subtypes of adrenocortical cancer ACC as an example,and it showed the practicability of COMSUC.This research demonstrates the application of drug repositioning prediction and cancer consensus molecular subtype identification based on transcriptional data and multi-omics data fusion.The main innovations of this research include the following three points.Firstly,we constructed a drug-induced gene co-expression network,which can help researchers to explore the mode of action of drugs.Secondly,the deep learning method was introduced into the drug repositioning,and improved the performance of drug-target interaction prediction.Thirdly,we developed a web server named COMSUC for cancer multi-omics data fusion and inconsistent clustering results of different cluster algorithm,which can accurately identify consensus molecular subtypes of cancer and related diagnostic biomarkers,and it can help make targeted medical diagnosis.