Molecular Subtypes Classification Of Colorectal Cancer And Prediction Of Colorectal Cancer-related Drugs

Colorectal cancer is the third most common cancer in the world,threatening human health.Because of its clinical heterogeneity,how to provide accurate and effective treatment for individual patients is always a difficult problem.Accurately discovering cancer subtypes is of great significance for the understanding of cancer,the treatment of cancer,and other practical clinical applications.Besides,the scarcity of anticancer drugs is another problem in the treatment of colorectal cancer.Because of the long cycle and low success rate of developing a new drug,it is more effective to excavate drugs that can be used to treat colorectal cancer from existing drugs.With the accumulation of biological data and the maturity of data mining technology,the use of advanced computing methods helps to solve the above problems.Overall,this paper has carried out a series of studies on subtypes of colorectal cancer and potential drugs for the treatment of colorectal cancer.The main contents are as follows:1.We propose a method called FCM-LightGBM for classifying molecular subtypes of colorectal cancer.First,we select differentially expressed genes from colorectal cancer patients.Then,FCM-based cluster analysis,clinical correlation analysis,pathway analysis,and survival analysis are applied to them.As a result,five distinct molecular subtypes of colorectal cancer can be identified.Combining differentially expressed gene data and subtypes,a LightGBM-based classifier is constructed.The experimental results show that the FCM-LightGBM based classification method can effectively predict subtypes for colorectal cancer patients.2.We propose a DGNLP-based method for predicting colorectal cancer-related drugs.To be specific,we take all cancer-related drugs as research objects,and build a drug-gene association network.Linear neighborhood similarity of drugs and genes can be obtained by taking drugs and genes as data points respectively and reconstructing them.On this basis,a label propagation algorithm is used to predict unknown drug-gene associations and unknown gene-drug associations.Then,we screen out the predicted associations related to the target genes of colorectal cancer,and select part of them for validation.What's more,we have combined with the results of the previous classification of colorectal cancer subtypes to search the drugs that may have therapeutic effects on specific subtypes.The experimental results show that the prediction accuracy of DGNLP is higher than the comparison methods,and most of the prediction results can be verified.3.We have established an online database of human colorectal cancer——HCCDB,which is user-friendly and easy to query.This database not only contains the drugs,pathways,and genes related to colorectal cancer,but also expands the biochemical information of existing drugs and genes.