Anticancer Drug Response Prediction Based On Deep Learning

Cancer is one of the most important diseases threatening human life and health.How to provide effective treatment for cancer patients has become an important topic in the medical field.The latest medical research results and clinical practice have showed that the effectiveness of existing anti-cancer drugs is highly dependent on the genomic characteristics of patients,which means that the efficacy of the same anti-cancer drugs may be very different for different patients even if they are suffering from the same cancer disease,since they usually have different genomic features.How to select the applicable anticancer drugs for different cancer patients is a challenge and frontier topic in the field of precision oncology.Accurate prediction of response between anticancer drugs and cancer cell lines is the premise and key for drug recommendation.The accuracy of drug response prediction model depends on the types of genomic data,data characterization methods,model design,training and evaluation,etc.Based on the above problems,an anticancer drug response prediction method based on deep convolutional neural network was proposed in this paper.Firstly,gene mutation,copy number variation and gene expression data from GDSC and COSMIC database were preprocessed and transformed into data presentation that could be learned by the model.Then the drug chemical structure files from Pub Chem database were processed and converted into molecular fingerprint features by Padel-descriptors software.Finally,response data were collected from 724 cancer cell lines and 212 anticancer drugs and each of cell lines contained three types of genomic data: expression,mutation,and copy number variation.In this paper,we design a hybrid convolutional neural network(CNN)to predict the responses of anti-cancer drugs,in which the network is constructed with two input CNN branches and two output CNN+FC(full connected)branches.For the two input branches,one is to extract the genomic feature from the input data of a cancer patient’s gene expression,mutation or copy number variations,and the other is to extract the molecular fingerprint feature from the drug to be used for curing the cancer.In addition,the attention mechanism,including both the channel and spatial attentions,is introduced to weight the two features according to their importance,the two weighted features are then fused into one vector and input to the two output branches.For the two output branches,one is to predict the IC50 values and the other is to predict the sensitivities of specific cancer cell line to anti-cancer drugs.For optimization of the model,this paper designs an end-to-end training method,and uses a joint loss function containing two kinds of loss functions to optimize the whole system.Extensive experiments and evaluations have been conducted on three types of genomic data(gene mutation,copy number variation and gene expression)in the work,in which the experimental results validate the effectiveness of the proposed method that outperforms the existing state of the art methods in terms of different performance indexes.In addition,this paper also developed a deep learning based anti-cancer drug response prediction prototype application system.The anticancer drug response prediction method and prototype application system proposed in this paper is expected to be applied for assisting targeted cancer therapy and assist doctors to recommend effective anticancer drugs to patients.