Research On Cell Line And Drug Response Prediction Based On Machine Learning

Genomic specificity can lead to different therapeutic effects of the same drug in different individuals suffering from the same disease.Therefore,personalized therapy based on individual genomic information is emerging,and predicting the different r esponses of specific cell lines to the same drug in different individuals is the core problem in personalized therapy.Validation of drug response through clinical experiments requires high requirements for experimental equipment and a large investment of time and money.With the development of high-throughput sequencing technology,it is of great practical significance to establish a model for preclinical drug response prediction based on biological data and computer algorithm.In this paper,based on the histological information of cell lines,chemical structure information of drugs and drug response information,two different drug response prediction models are established using machine learning methods,and the main work is as follows:(1)In this study,a Drug Response Prediction model based on Maximum flow and Random Forest is proposed.It obtains cell line gene expression data,drug chemical structure data,and known cell line-drug response data from a database.Firstly,the data are processed for similarity metric to obtain the initial similarity network.Then,the maximum flow algorithm is introduced as a feature selection strategy to obtain the features of cell lines and drugs,respectively,which are concatenated to obtain the cell line-drug pair feature vector.Finally,the cell line-drug classification is performed by random forest.(2)In this study,a Drug Response Prediction model based on maximum flow and Convolutional Neural Networks is proposed.It obtains multi-omics data of cell lines from database analysis,including gene expression,gene mutation,copy number variation,drug chemical structure data and known cell line-drug response data.Then,a similarity network is constructed on this basis,and the maximum flow algorithm is used to extract the features of cell lines and drugs.Finally,a 9-layer convolutional neural network is built to predict the relationships of cell line-drug pairs.(3)In this study,Mflow＿RFDRP,Mflow＿CNNDRP and existing similar drug response prediction algorithms were tested in comparison from different perspectives.Compared with comparable existing calculations,broad test comes about appear that Mflow＿RFDRP and Mflow＿CNNDRP have superior expectation execution.The prediction performance of Mflow＿RFDRP and Mflow＿CNNDRP is comparable when using only single-omics data such as gene expression,but the Mflow＿RFDRP runtime is much shorter than that of Mflow＿CNNDRP.In contrast,the Mflow＿CNNDRP model performs better in prediction performance when dealing with multi-omics data,but it also takes longer time.