*2.9. Implementation and Code Availability*

All the neural networks below were implemented in Keras (2.0.9) with Tensorflow (1.4.0) on a GPU server (NVIDIA Telsa K40 GPU). The choice of the parameters for gradient boosting followed TargetFinder (https://github.com/shwhalen/targetfinder), and we chose the # of trees through the validation dataset. Gradient boosting was implemented with Python Sklearn. The code is available at https://github.com/menglix/EPI.
