JavaDL: a Java-based Deep Learning Tool to Predict Drug Responses

摘要: Motivation: Accurate prediction of drug response in each patient is the holy grail personalized medicine. Recently, deep learning techniques have been witnessed with revival a variety areas such as image processing and genomic data analysis, they will be useful for coming age big analysis pharmaceutical research chemogenomic applications. This provides us an impetus to develop novel platform accurately reliably predict cancer different treatments. Results: In this study, we describe Java-based implementation neural network (DNN) method, termed JavaDL, responses drugs solely based on their chemical features. To end, devised cost function by adding regularization term which suppresses overfitting. We also adopted 揺arly stopping?strategy further reduce overfit improve accuracy robustness our models. Currently software has integrated genetic algorithm-based variable selection approach implemented part JavaDL package. evaluate program, compared it several machine programs including SVM kNN. observed that either significantly outperforms other methods model building or obtains better results handling analysis. Finally, was employed highly aggressive triple-negative breast cell lines, showed robust accurate predictions r2 high 0.80.