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

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