Prediction Of Synergistic Effects Of Anticancer Drug Combinations Based On Deep Learnin

Cancer is often caused by polygenic abnormalities.In cancer chemotherapy,the traditional monotherapy has been proved to have obvious defects in cancer treatment.For example,monotherapy is easy to produce large toxic and side effects and toxicity.Compared with single drug therapy,combination therapy can not only improve drug efficacy,but also reduce drug resistance and reduce drug toxicity.In addition,the method of drug combination screening can also reposition the drugs already on the market,which is of great significance to reduce the economic burden of new drug discovery.In recent years,the number of drug combinations has increased exponentially with the increase of the number of drugs,and due to the heterogeneity of cancer cells,it is also necessary to test the effect of combined drugs on different types of cancer cells in cell experiments.However,it is difficult for current equipment to screen drug combinations quickly and with high throughput.To solve the above problems,this paper proposed two schemes of drug combinations synergy prediction based on deep learning,which can predict the synergy score of drug combinations in regression.In the first method,we use one-hot coding,drug and cell line feature descriptors to represent the features of combined drug prediction,use mol2 vec model to transform the SMILES of drug molecules into vectors,and proposed a factorization machine based deep neural network(DNN-FM),which combined the structure of Factorization Machine and Deep Neural Network to predict the synergy of drug combinations.Compared with traditional machine learning methods,DNN-FM has better performance in processing complex data.The comprehensive experimental results show that DNN-FM has the characteristics of high efficiency and high precision in the prediction of drug combination synergy.Due to the first method is not an end-to-end model,it is difficult to integrate the model in modularity.Therefore,in the second method,we propose an end-to-end deep learning model based on convolutional neural network to predict the synergy of drug combination,and use different databases to evaluate the multi-dimensional performance of the model.The results show that compared with other models,this model has excellent classification performance and regression prediction performance.In conclusion,this study is a new attempt in the research of drug combination screening.After different feature representations of drug combination information data and cancer cell line data,the prediction of drug combination synergy with better performance is realized.The root mean square errors of DNN-FM model and end-toend model based on convolutional neural network reach 7.600 and 4.901 respectively.The model proposed in this study can provide a reference for the primary screening of large-scale and high-throughput anticancer drugs.