Research On Prediction Of Drug Responses Of Cancer Cells Based On Maximum Flow

Studying the response of cancer cell lines to anticancer drugs is the key to understanding cancer biology and developing new anticancer therapies,which plays a vital role in cancer treatment.The traditional method of studying cancer cell line-drug response is based on clinical experiments,which is a time-consuming and costly process,and its clinical relevance is controversial.In search of a low-investment,short-cycle and preclinical approach,researchers began to use known biological data to model drug response sensitivity.Researches have shown that the effective integration of multiple biomolecular information can greatly improve the accuracy of cell line-drug response prediction.This article mainly studied the prediction of cell line-drug response by effectively integrating multi-source data and establishing a suitable model.The work done is as follows:(1)We proposed MIF2DRP(Using Maximum Information Flow on Heterogeneous Network For Drug Response Prediction)model to predict cell line-drug response.First,we downloaded the cell line,drug and gene data from multiple databases.After the similarity measurement and the preprocessing of these data,a three-layer heterogeneous network containing multi-source information was integrated.Then,it was converted to a flow network.For each drug,the maximum information flow between drug node and cell line node was calculated by using the maximum flow algorithm.This value was used to measure the relationship between cell and drug.Finally,by setting the threshold,the cell line's response to the drug was classified as sensitive or resistant,and the final prediction result was obtained.Experimental results showed that the overall performance of MIF2 DRP is the best when compared with the other two advanced prediction models.(2)We proposed RFMMIF(A Random Forest Method Based on Maximum Information Flow For Drug Response Prediction)model to predict cell line-drug response.After preprocessing the known characteristic information and relationship information of drug and cell in database and literature,a drug relationship network and a cell line relationship network were obtained.Then the importa nce weights of the nodes in the two networks were calculated by the maximum flow method,and the known drug characteristics and cell line characteristics were weighted according to the weights.We obtained the feature vector of the cell line-drug pair.Finally,after feature selection procedure,a random forest classification model was used to classify the cell line-drug pair to complete the prediction work.The experimental results showed that our method had good prediction performance,and was also well in predicting the unknown cell line-drug response.